{"schema_version":1,"snapshot_utc":"2026-11-12T12:42:56.402836+00:00","reference_cell":"cap_10000_n100","primary_analysis_view":"strict","models":{"AUTOGLUON":{"id":"AUTOGLUON","display":"AutoGluon","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"AutoGluon automates tabular learning by training multiple models and combining their predictions in ensembles. Here it runs its native extreme preset within a one-hour budget as a separate AutoML reference.","tuning":null,"category":"AutoML","color":"#111827","regular_max_features":null,"training_data_overlap":false},"CAT":{"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},"DUMMY":{"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},"GBM":{"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},"KNN":{"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},"LR":{"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},"MITRA":{"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},"NN_TORCH":{"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},"REALMLP":{"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},"REALTABPFN-V2":{"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},"REALTABPFN-V2.5":{"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},"RF":{"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},"TABDPT":{"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},"TABFM":{"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},"TABICL":{"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},"TABM":{"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},"TABPFN-V3":{"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},"TABPFN-WIDE":{"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},"TABPFN-WIDE-5K-NE3":{"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},"XGB":{"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},"XT":{"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false}},"cell_options":[{"id":"cap_2000_n20","label":"p=2,000, n=20","feature_cap":2000,"n_train":20},{"id":"cap_2000_n50","label":"p=2,000, n=50","feature_cap":2000,"n_train":50},{"id":"cap_2000_n100","label":"p=2,000, n=100","feature_cap":2000,"n_train":100},{"id":"cap_2000_n200","label":"p=2,000, n=200","feature_cap":2000,"n_train":200},{"id":"cap_2000_n500","label":"p=2,000, n=500","feature_cap":2000,"n_train":500},{"id":"cap_2000_n1000","label":"p=2,000, n=1,000","feature_cap":2000,"n_train":1000},{"id":"cap_2000","label":"p=2,000, n=all","feature_cap":2000,"n_train":null},{"id":"cap_10000_n20","label":"p=10,000, n=20","feature_cap":10000,"n_train":20},{"id":"cap_10000_n50","label":"p=10,000, n=50","feature_cap":10000,"n_train":50},{"id":"cap_10000_n100","label":"p=10,000, n=100","feature_cap":10000,"n_train":100},{"id":"cap_10000_n200","label":"p=10,000, n=200","feature_cap":10000,"n_train":200},{"id":"cap_10000_n500","label":"p=10,000, n=500","feature_cap":10000,"n_train":500},{"id":"cap_10000_n1000","label":"p=10,000, n=1,000","feature_cap":10000,"n_train":1000},{"id":"cap_10000","label":"p=10,000, n=all","feature_cap":10000,"n_train":null},{"id":"cap_25000_n20","label":"p=25,000, n=20","feature_cap":25000,"n_train":20},{"id":"cap_25000_n50","label":"p=25,000, n=50","feature_cap":25000,"n_train":50},{"id":"cap_25000_n100","label":"p=25,000, n=100","feature_cap":25000,"n_train":100},{"id":"cap_25000_n200","label":"p=25,000, n=200","feature_cap":25000,"n_train":200},{"id":"cap_25000_n500","label":"p=25,000, n=500","feature_cap":25000,"n_train":500},{"id":"cap_25000_n1000","label":"p=25,000, n=1,000","feature_cap":25000,"n_train":1000},{"id":"cap_25000","label":"p=25,000, n=all","feature_cap":25000,"n_train":null},{"id":"cap_full_n20","label":"p=all, n=20","feature_cap":null,"n_train":20},{"id":"cap_full_n50","label":"p=all, n=50","feature_cap":null,"n_train":50},{"id":"cap_full_n100","label":"p=all, n=100","feature_cap":null,"n_train":100},{"id":"cap_full_n200","label":"p=all, n=200","feature_cap":null,"n_train":200},{"id":"cap_full_n500","label":"p=all, n=500","feature_cap":null,"n_train":500},{"id":"cap_full_n1000","label":"p=all, n=1,000","feature_cap":null,"n_train":1000},{"id":"cap_full","label":"p=all, n=all","feature_cap":null,"n_train":null}],"reference":[{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1155,"Elo_lo":1098,"Elo_hi":1211,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false,"f1_macro":0.713391503894652,"rmse":1.742894159922996},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1071,"Elo_lo":1016,"Elo_hi":1124,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false,"f1_macro":0.6854581607208678,"rmse":1.8372622165329882},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1063,"Elo_lo":994,"Elo_hi":1126,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.7205882559484305,"rmse":1.955818026331612},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1045,"Elo_lo":988,"Elo_hi":1099,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.7008845111802758,"rmse":2.115544869242223},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1038,"Elo_lo":952,"Elo_hi":1114,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.674900372788093,"rmse":null},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"AUTOGLUON","Elo":1036,"Elo_lo":984,"Elo_hi":1085,"n_targets":42,"id":"AUTOGLUON","display":"AutoGluon","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"AutoGluon automates tabular learning by training multiple models and combining their predictions in ensembles. Here it runs its native extreme preset within a one-hour budget as a separate AutoML reference.","tuning":null,"category":"AutoML","color":"#111827","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6945642015955917,"rmse":2.0725846227782996},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1033,"Elo_lo":979,"Elo_hi":1087,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false,"f1_macro":0.6646341040192109,"rmse":1.8178014682508676},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1031,"Elo_lo":976,"Elo_hi":1085,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true,"f1_macro":0.6939871342451706,"rmse":2.0947178687450423},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1019,"Elo_lo":932,"Elo_hi":1095,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6681470028760327,"rmse":null},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1010,"Elo_lo":963,"Elo_hi":1059,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6812042125308531,"rmse":2.015024780475652},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6658157947879919,"rmse":1.9304502775369867},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":994,"Elo_lo":929,"Elo_hi":1056,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false,"f1_macro":0.6331429574772196,"rmse":1.791143875950348},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":993,"Elo_lo":923,"Elo_hi":1059,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.69916754760155,"rmse":2.1705489086171754},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":973,"Elo_lo":937,"Elo_hi":1011,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6445308419027305,"rmse":1.9776414097758406},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":973,"Elo_lo":945,"Elo_hi":999,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6442367032136745,"rmse":1.998151006713182},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":956,"Elo_lo":900,"Elo_hi":1007,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false,"f1_macro":0.6472922420093705,"rmse":1.902521315844419},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":906,"Elo_lo":851,"Elo_hi":961,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6386446187985723,"rmse":1.8806553715934375},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":867,"Elo_lo":820,"Elo_hi":913,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6349965657930864,"rmse":2.109612748031864},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":865,"Elo_lo":810,"Elo_hi":920,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6290497675098728,"rmse":2.183340886639696},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":744,"Elo_lo":638,"Elo_hi":837,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false,"f1_macro":0.3601253990827079,"rmse":2.167685045473319},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":522,"Elo_lo":457,"Elo_hi":583,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.32326853560954805,"rmse":2.2366634323898165}],"cell_elo":[{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1069,"Elo_lo":1020,"Elo_hi":1117,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1040,"Elo_lo":982,"Elo_hi":1096,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1025,"Elo_lo":977,"Elo_hi":1073,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1019,"Elo_lo":975,"Elo_hi":1064,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1010,"Elo_lo":962,"Elo_hi":1062,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1007,"Elo_lo":937,"Elo_hi":1078,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":1000,"Elo_lo":971,"Elo_hi":1026,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":987,"Elo_lo":948,"Elo_hi":1023,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":970,"Elo_lo":898,"Elo_hi":1041,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":957,"Elo_lo":886,"Elo_hi":1027,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":940,"Elo_lo":883,"Elo_hi":998,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":933,"Elo_lo":883,"Elo_hi":981,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":907,"Elo_lo":856,"Elo_hi":956,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":884,"Elo_lo":819,"Elo_hi":948,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":882,"Elo_lo":825,"Elo_hi":939,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":875,"Elo_lo":791,"Elo_hi":955,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":826,"Elo_lo":778,"Elo_hi":870,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":706,"Elo_lo":630,"Elo_hi":775,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":706,"Elo_lo":629,"Elo_hi":778,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1120,"Elo_lo":1066,"Elo_hi":1173,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1080,"Elo_lo":1024,"Elo_hi":1132,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1079,"Elo_lo":1024,"Elo_hi":1133,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1062,"Elo_lo":1007,"Elo_hi":1116,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1015,"Elo_lo":966,"Elo_hi":1067,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1012,"Elo_lo":938,"Elo_hi":1092,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1008,"Elo_lo":922,"Elo_hi":1094,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1007,"Elo_lo":951,"Elo_hi":1060,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1000,"Elo_lo":952,"Elo_hi":1045,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":986,"Elo_lo":963,"Elo_hi":1011,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":985,"Elo_lo":933,"Elo_hi":1033,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":968,"Elo_lo":914,"Elo_hi":1022,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":963,"Elo_lo":899,"Elo_hi":1027,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":960,"Elo_lo":925,"Elo_hi":995,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":919,"Elo_lo":869,"Elo_hi":970,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":841,"Elo_lo":794,"Elo_hi":883,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":834,"Elo_lo":752,"Elo_hi":910,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":785,"Elo_lo":727,"Elo_hi":841,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":636,"Elo_lo":566,"Elo_hi":701,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1154,"Elo_lo":1092,"Elo_hi":1219,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1112,"Elo_lo":1052,"Elo_hi":1172,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1100,"Elo_lo":1036,"Elo_hi":1166,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1094,"Elo_lo":1034,"Elo_hi":1153,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1090,"Elo_lo":1022,"Elo_hi":1161,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1088,"Elo_lo":1018,"Elo_hi":1154,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1083,"Elo_lo":1018,"Elo_hi":1149,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1066,"Elo_lo":974,"Elo_hi":1152,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1048,"Elo_lo":1002,"Elo_hi":1102,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1046,"Elo_lo":996,"Elo_hi":1098,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1038,"Elo_lo":952,"Elo_hi":1113,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1004,"Elo_lo":932,"Elo_hi":1069,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":988,"Elo_lo":966,"Elo_hi":1009,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":972,"Elo_lo":941,"Elo_hi":1002,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":916,"Elo_lo":860,"Elo_hi":972,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":876,"Elo_lo":824,"Elo_hi":923,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":869,"Elo_lo":806,"Elo_hi":929,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":829,"Elo_lo":740,"Elo_hi":909,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":556,"Elo_lo":487,"Elo_hi":614,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1197,"Elo_lo":1130,"Elo_hi":1269,"n_targets":34,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1152,"Elo_lo":1086,"Elo_hi":1228,"n_targets":34,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1130,"Elo_lo":1043,"Elo_hi":1219,"n_targets":34,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1122,"Elo_lo":1048,"Elo_hi":1196,"n_targets":34,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1118,"Elo_lo":1039,"Elo_hi":1199,"n_targets":34,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1107,"Elo_lo":1037,"Elo_hi":1172,"n_targets":34,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1104,"Elo_lo":1015,"Elo_hi":1196,"n_targets":22,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1098,"Elo_lo":1005,"Elo_hi":1183,"n_targets":22,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1083,"Elo_lo":1015,"Elo_hi":1151,"n_targets":34,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1040,"Elo_lo":949,"Elo_hi":1131,"n_targets":34,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1024,"Elo_lo":959,"Elo_hi":1090,"n_targets":34,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1016,"Elo_lo":942,"Elo_hi":1089,"n_targets":34,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":34,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":993,"Elo_lo":971,"Elo_hi":1014,"n_targets":34,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":987,"Elo_lo":940,"Elo_hi":1033,"n_targets":34,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":966,"Elo_lo":910,"Elo_hi":1016,"n_targets":34,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":919,"Elo_lo":867,"Elo_hi":975,"n_targets":34,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":886,"Elo_lo":777,"Elo_hi":987,"n_targets":34,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":842,"Elo_lo":770,"Elo_hi":914,"n_targets":34,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":498,"Elo_lo":427,"Elo_hi":570,"n_targets":34,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1187,"Elo_lo":1104,"Elo_hi":1276,"n_targets":20,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1135,"Elo_lo":1036,"Elo_hi":1245,"n_targets":20,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1134,"Elo_lo":1021,"Elo_hi":1256,"n_targets":12,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1089,"Elo_lo":1015,"Elo_hi":1168,"n_targets":20,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1088,"Elo_lo":986,"Elo_hi":1204,"n_targets":20,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1063,"Elo_lo":913,"Elo_hi":1226,"n_targets":20,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1049,"Elo_lo":967,"Elo_hi":1138,"n_targets":20,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1047,"Elo_lo":946,"Elo_hi":1145,"n_targets":20,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1030,"Elo_lo":935,"Elo_hi":1134,"n_targets":20,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1030,"Elo_lo":882,"Elo_hi":1180,"n_targets":12,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1004,"Elo_lo":952,"Elo_hi":1056,"n_targets":20,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":20,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":1000,"Elo_lo":967,"Elo_hi":1032,"n_targets":20,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":980,"Elo_lo":914,"Elo_hi":1041,"n_targets":20,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":922,"Elo_lo":811,"Elo_hi":1039,"n_targets":20,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":913,"Elo_lo":856,"Elo_hi":981,"n_targets":20,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":871,"Elo_lo":722,"Elo_hi":1008,"n_targets":20,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":783,"Elo_lo":595,"Elo_hi":948,"n_targets":20,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":757,"Elo_lo":644,"Elo_hi":859,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":448,"Elo_lo":339,"Elo_hi":550,"n_targets":20,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1143,"Elo_lo":974,"Elo_hi":1322,"n_targets":15,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1143,"Elo_lo":987,"Elo_hi":1306,"n_targets":8,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1135,"Elo_lo":998,"Elo_hi":1291,"n_targets":15,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1087,"Elo_lo":955,"Elo_hi":1222,"n_targets":15,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1073,"Elo_lo":936,"Elo_hi":1198,"n_targets":15,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1034,"Elo_lo":964,"Elo_hi":1098,"n_targets":15,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1017,"Elo_lo":811,"Elo_hi":1225,"n_targets":15,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1011,"Elo_lo":857,"Elo_hi":1179,"n_targets":15,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1006,"Elo_lo":934,"Elo_hi":1083,"n_targets":15,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1001,"Elo_lo":849,"Elo_hi":1147,"n_targets":15,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":15,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":983,"Elo_lo":940,"Elo_hi":1018,"n_targets":15,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":965,"Elo_lo":762,"Elo_hi":1171,"n_targets":8,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":955,"Elo_lo":865,"Elo_hi":1043,"n_targets":15,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":859,"Elo_lo":683,"Elo_hi":1014,"n_targets":15,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":751,"Elo_lo":591,"Elo_hi":892,"n_targets":15,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":733,"Elo_lo":431,"Elo_hi":955,"n_targets":15,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":696,"Elo_lo":551,"Elo_hi":816,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":557,"Elo_lo":249,"Elo_hi":806,"n_targets":15,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":403,"Elo_lo":220,"Elo_hi":546,"n_targets":15,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1145,"Elo_lo":1084,"Elo_hi":1210,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1125,"Elo_lo":1070,"Elo_hi":1179,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1123,"Elo_lo":1068,"Elo_hi":1187,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1104,"Elo_lo":1028,"Elo_hi":1177,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1101,"Elo_lo":1054,"Elo_hi":1150,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1081,"Elo_lo":1019,"Elo_hi":1140,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1068,"Elo_lo":1001,"Elo_hi":1136,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1057,"Elo_lo":1012,"Elo_hi":1106,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1050,"Elo_lo":991,"Elo_hi":1106,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1049,"Elo_lo":947,"Elo_hi":1145,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1043,"Elo_lo":1009,"Elo_hi":1079,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1037,"Elo_lo":961,"Elo_hi":1108,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":993,"Elo_lo":951,"Elo_hi":1036,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":990,"Elo_lo":964,"Elo_hi":1012,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":989,"Elo_lo":910,"Elo_hi":1063,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":988,"Elo_lo":904,"Elo_hi":1067,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":881,"Elo_lo":833,"Elo_hi":929,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":623,"Elo_lo":485,"Elo_hi":732,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":494,"Elo_lo":426,"Elo_hi":550,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1049,"Elo_lo":991,"Elo_hi":1108,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1047,"Elo_lo":997,"Elo_hi":1091,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1045,"Elo_lo":997,"Elo_hi":1090,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1011,"Elo_lo":944,"Elo_hi":1078,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":1007,"Elo_lo":986,"Elo_hi":1028,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":996,"Elo_lo":961,"Elo_hi":1026,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":986,"Elo_lo":938,"Elo_hi":1037,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":972,"Elo_lo":902,"Elo_hi":1037,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":966,"Elo_lo":903,"Elo_hi":1026,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":956,"Elo_lo":910,"Elo_hi":997,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":954,"Elo_lo":885,"Elo_hi":1017,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":944,"Elo_lo":897,"Elo_hi":987,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":936,"Elo_lo":889,"Elo_hi":983,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":892,"Elo_lo":848,"Elo_hi":937,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":888,"Elo_lo":835,"Elo_hi":941,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":825,"Elo_lo":781,"Elo_hi":870,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":750,"Elo_lo":666,"Elo_hi":831,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":702,"Elo_lo":626,"Elo_hi":772,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":697,"Elo_lo":622,"Elo_hi":767,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1132,"Elo_lo":1079,"Elo_hi":1182,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1081,"Elo_lo":1028,"Elo_hi":1133,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1081,"Elo_lo":1025,"Elo_hi":1132,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1034,"Elo_lo":947,"Elo_hi":1120,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1025,"Elo_lo":971,"Elo_hi":1078,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1022,"Elo_lo":964,"Elo_hi":1080,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1021,"Elo_lo":944,"Elo_hi":1101,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1006,"Elo_lo":946,"Elo_hi":1068,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":989,"Elo_lo":928,"Elo_hi":1054,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":966,"Elo_hi":1010,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":971,"Elo_lo":916,"Elo_hi":1026,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":970,"Elo_lo":927,"Elo_hi":1011,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":965,"Elo_lo":928,"Elo_hi":1002,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":950,"Elo_lo":897,"Elo_hi":1001,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":927,"Elo_lo":871,"Elo_hi":979,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":835,"Elo_lo":795,"Elo_hi":877,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":824,"Elo_lo":771,"Elo_hi":873,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":763,"Elo_lo":677,"Elo_hi":840,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":625,"Elo_lo":556,"Elo_hi":690,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1155,"Elo_lo":1098,"Elo_hi":1211,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1071,"Elo_lo":1016,"Elo_hi":1124,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1063,"Elo_lo":994,"Elo_hi":1126,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1045,"Elo_lo":988,"Elo_hi":1099,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1038,"Elo_lo":952,"Elo_hi":1114,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"AUTOGLUON","Elo":1036,"Elo_lo":984,"Elo_hi":1085,"n_targets":42,"id":"AUTOGLUON","display":"AutoGluon","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"AutoGluon automates tabular learning by training multiple models and combining their predictions in ensembles. Here it runs its native extreme preset within a one-hour budget as a separate AutoML reference.","tuning":null,"category":"AutoML","color":"#111827","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1033,"Elo_lo":979,"Elo_hi":1087,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1031,"Elo_lo":976,"Elo_hi":1085,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1019,"Elo_lo":932,"Elo_hi":1095,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1010,"Elo_lo":963,"Elo_hi":1059,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":994,"Elo_lo":929,"Elo_hi":1056,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":993,"Elo_lo":923,"Elo_hi":1059,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":973,"Elo_lo":937,"Elo_hi":1011,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":973,"Elo_lo":945,"Elo_hi":999,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":956,"Elo_lo":900,"Elo_hi":1007,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":906,"Elo_lo":851,"Elo_hi":961,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":867,"Elo_lo":820,"Elo_hi":913,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":865,"Elo_lo":810,"Elo_hi":920,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":744,"Elo_lo":638,"Elo_hi":837,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":522,"Elo_lo":457,"Elo_hi":583,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1172,"Elo_lo":1110,"Elo_hi":1234,"n_targets":34,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1123,"Elo_lo":1054,"Elo_hi":1195,"n_targets":34,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1106,"Elo_lo":1008,"Elo_hi":1206,"n_targets":22,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1069,"Elo_lo":991,"Elo_hi":1139,"n_targets":34,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1068,"Elo_lo":998,"Elo_hi":1135,"n_targets":34,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1063,"Elo_lo":980,"Elo_hi":1146,"n_targets":34,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1062,"Elo_lo":960,"Elo_hi":1165,"n_targets":22,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1061,"Elo_lo":997,"Elo_hi":1131,"n_targets":34,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1050,"Elo_lo":965,"Elo_hi":1139,"n_targets":34,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1000,"Elo_lo":958,"Elo_hi":1044,"n_targets":34,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":34,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":994,"Elo_lo":930,"Elo_hi":1058,"n_targets":34,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":984,"Elo_lo":910,"Elo_hi":1062,"n_targets":34,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":977,"Elo_lo":922,"Elo_hi":1033,"n_targets":34,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":977,"Elo_lo":954,"Elo_hi":998,"n_targets":34,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":954,"Elo_lo":879,"Elo_hi":1020,"n_targets":34,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":915,"Elo_lo":853,"Elo_hi":975,"n_targets":34,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":833,"Elo_lo":767,"Elo_hi":895,"n_targets":34,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":817,"Elo_lo":693,"Elo_hi":930,"n_targets":34,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":482,"Elo_lo":414,"Elo_hi":552,"n_targets":34,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1153,"Elo_lo":1066,"Elo_hi":1243,"n_targets":20,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1123,"Elo_lo":1044,"Elo_hi":1218,"n_targets":20,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1094,"Elo_lo":976,"Elo_hi":1215,"n_targets":12,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1066,"Elo_lo":987,"Elo_hi":1149,"n_targets":20,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1065,"Elo_lo":990,"Elo_hi":1144,"n_targets":20,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1045,"Elo_lo":938,"Elo_hi":1154,"n_targets":20,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1029,"Elo_lo":929,"Elo_hi":1123,"n_targets":20,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1020,"Elo_lo":955,"Elo_hi":1086,"n_targets":20,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1012,"Elo_lo":865,"Elo_hi":1166,"n_targets":12,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":20,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":988,"Elo_lo":956,"Elo_hi":1016,"n_targets":20,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":984,"Elo_lo":919,"Elo_hi":1050,"n_targets":20,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":942,"Elo_lo":875,"Elo_hi":1010,"n_targets":20,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":933,"Elo_lo":829,"Elo_hi":1039,"n_targets":20,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":908,"Elo_lo":749,"Elo_hi":1059,"n_targets":20,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":870,"Elo_lo":771,"Elo_hi":968,"n_targets":20,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":806,"Elo_lo":665,"Elo_hi":959,"n_targets":20,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":765,"Elo_lo":578,"Elo_hi":934,"n_targets":20,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":753,"Elo_lo":637,"Elo_hi":849,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":457,"Elo_lo":349,"Elo_hi":557,"n_targets":20,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1112,"Elo_lo":964,"Elo_hi":1277,"n_targets":15,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1112,"Elo_lo":919,"Elo_hi":1300,"n_targets":8,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1100,"Elo_lo":950,"Elo_hi":1258,"n_targets":15,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1063,"Elo_lo":934,"Elo_hi":1193,"n_targets":15,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1044,"Elo_lo":895,"Elo_hi":1177,"n_targets":15,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1015,"Elo_lo":929,"Elo_hi":1094,"n_targets":15,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":15,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":997,"Elo_lo":903,"Elo_hi":1094,"n_targets":15,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":992,"Elo_lo":960,"Elo_hi":1018,"n_targets":15,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":990,"Elo_lo":833,"Elo_hi":1143,"n_targets":15,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":988,"Elo_lo":822,"Elo_hi":1158,"n_targets":15,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":946,"Elo_lo":856,"Elo_hi":1036,"n_targets":15,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":934,"Elo_lo":698,"Elo_hi":1163,"n_targets":8,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":864,"Elo_lo":694,"Elo_hi":1043,"n_targets":15,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":834,"Elo_lo":649,"Elo_hi":997,"n_targets":15,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":739,"Elo_lo":427,"Elo_hi":967,"n_targets":15,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":686,"Elo_lo":542,"Elo_hi":813,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":676,"Elo_lo":529,"Elo_hi":817,"n_targets":15,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":551,"Elo_lo":250,"Elo_hi":798,"n_targets":15,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":405,"Elo_lo":219,"Elo_hi":548,"n_targets":15,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1133,"Elo_lo":1071,"Elo_hi":1198,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1127,"Elo_lo":1070,"Elo_hi":1183,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1116,"Elo_lo":1059,"Elo_hi":1182,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1092,"Elo_lo":1012,"Elo_hi":1170,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1087,"Elo_lo":1025,"Elo_hi":1146,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1083,"Elo_lo":1032,"Elo_hi":1135,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1068,"Elo_lo":999,"Elo_hi":1133,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1063,"Elo_lo":1020,"Elo_hi":1108,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1059,"Elo_lo":1018,"Elo_hi":1103,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1027,"Elo_lo":960,"Elo_hi":1095,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1019,"Elo_lo":942,"Elo_hi":1097,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1004,"Elo_lo":916,"Elo_hi":1083,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":1003,"Elo_lo":957,"Elo_hi":1051,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":978,"Elo_lo":954,"Elo_hi":1003,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":912,"Elo_lo":813,"Elo_hi":1003,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":880,"Elo_lo":834,"Elo_hi":929,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":785,"Elo_lo":714,"Elo_hi":852,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":580,"Elo_lo":443,"Elo_hi":698,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":481,"Elo_lo":412,"Elo_hi":540,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1056,"Elo_lo":1004,"Elo_hi":1105,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1031,"Elo_lo":974,"Elo_hi":1089,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1027,"Elo_lo":977,"Elo_hi":1076,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":967,"Elo_hi":1011,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":982,"Elo_lo":915,"Elo_hi":1050,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":970,"Elo_lo":901,"Elo_hi":1037,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":968,"Elo_lo":914,"Elo_hi":1023,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":965,"Elo_lo":929,"Elo_hi":1001,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":959,"Elo_lo":897,"Elo_hi":1020,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":941,"Elo_lo":869,"Elo_hi":1011,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":929,"Elo_lo":884,"Elo_hi":973,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":921,"Elo_lo":868,"Elo_hi":969,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":908,"Elo_lo":855,"Elo_hi":959,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":871,"Elo_lo":826,"Elo_hi":916,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":865,"Elo_lo":808,"Elo_hi":924,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":793,"Elo_lo":748,"Elo_hi":836,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":736,"Elo_lo":652,"Elo_hi":818,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":687,"Elo_lo":610,"Elo_hi":758,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":685,"Elo_lo":608,"Elo_hi":755,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1140,"Elo_lo":1088,"Elo_hi":1188,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1077,"Elo_lo":1020,"Elo_hi":1131,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1077,"Elo_lo":1022,"Elo_hi":1128,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1023,"Elo_lo":972,"Elo_hi":1078,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1014,"Elo_lo":929,"Elo_hi":1097,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":993,"Elo_lo":921,"Elo_hi":1068,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":986,"Elo_lo":966,"Elo_hi":1006,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":974,"Elo_lo":916,"Elo_hi":1034,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":970,"Elo_lo":924,"Elo_hi":1014,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":962,"Elo_lo":908,"Elo_hi":1020,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":950,"Elo_lo":915,"Elo_hi":987,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":935,"Elo_lo":884,"Elo_hi":990,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":928,"Elo_lo":873,"Elo_hi":982,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":926,"Elo_lo":873,"Elo_hi":979,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":925,"Elo_lo":866,"Elo_hi":980,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":841,"Elo_lo":796,"Elo_hi":887,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":832,"Elo_lo":782,"Elo_hi":880,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":752,"Elo_lo":665,"Elo_hi":832,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":618,"Elo_lo":554,"Elo_hi":682,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1174,"Elo_lo":1115,"Elo_hi":1236,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1082,"Elo_lo":1029,"Elo_hi":1134,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1060,"Elo_lo":992,"Elo_hi":1120,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1044,"Elo_lo":986,"Elo_hi":1106,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1041,"Elo_lo":986,"Elo_hi":1096,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1041,"Elo_lo":955,"Elo_hi":1117,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1028,"Elo_lo":939,"Elo_hi":1109,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1017,"Elo_lo":945,"Elo_hi":1084,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1016,"Elo_lo":965,"Elo_hi":1066,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":983,"Elo_lo":948,"Elo_hi":1019,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":982,"Elo_lo":965,"Elo_hi":1001,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":978,"Elo_lo":910,"Elo_hi":1040,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":973,"Elo_lo":908,"Elo_hi":1038,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":929,"Elo_lo":874,"Elo_hi":980,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":914,"Elo_lo":853,"Elo_hi":977,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":881,"Elo_lo":820,"Elo_hi":936,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":876,"Elo_lo":830,"Elo_hi":920,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":755,"Elo_lo":652,"Elo_hi":842,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":549,"Elo_lo":486,"Elo_hi":604,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1189,"Elo_lo":1132,"Elo_hi":1247,"n_targets":34,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1119,"Elo_lo":1056,"Elo_hi":1186,"n_targets":34,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1085,"Elo_lo":997,"Elo_hi":1171,"n_targets":22,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1068,"Elo_lo":1005,"Elo_hi":1129,"n_targets":34,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1063,"Elo_lo":1006,"Elo_hi":1116,"n_targets":34,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1053,"Elo_lo":958,"Elo_hi":1143,"n_targets":22,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1051,"Elo_lo":971,"Elo_hi":1128,"n_targets":34,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1025,"Elo_lo":945,"Elo_hi":1099,"n_targets":34,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1010,"Elo_lo":939,"Elo_hi":1077,"n_targets":34,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1001,"Elo_lo":966,"Elo_hi":1039,"n_targets":34,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":34,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":998,"Elo_lo":933,"Elo_hi":1056,"n_targets":34,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":992,"Elo_lo":937,"Elo_hi":1043,"n_targets":34,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":968,"Elo_hi":1009,"n_targets":34,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":940,"Elo_lo":851,"Elo_hi":1034,"n_targets":34,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":917,"Elo_lo":861,"Elo_hi":974,"n_targets":34,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":904,"Elo_lo":826,"Elo_hi":974,"n_targets":34,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":845,"Elo_lo":779,"Elo_hi":908,"n_targets":34,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":823,"Elo_lo":708,"Elo_hi":930,"n_targets":34,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":522,"Elo_lo":463,"Elo_hi":579,"n_targets":34,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1157,"Elo_lo":1065,"Elo_hi":1247,"n_targets":20,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1112,"Elo_lo":1027,"Elo_hi":1214,"n_targets":20,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1078,"Elo_lo":976,"Elo_hi":1177,"n_targets":12,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1077,"Elo_lo":997,"Elo_hi":1162,"n_targets":20,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1070,"Elo_lo":989,"Elo_hi":1150,"n_targets":20,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1031,"Elo_lo":932,"Elo_hi":1132,"n_targets":20,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1028,"Elo_lo":916,"Elo_hi":1136,"n_targets":20,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1027,"Elo_lo":961,"Elo_hi":1097,"n_targets":20,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":20,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":992,"Elo_lo":832,"Elo_hi":1149,"n_targets":12,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":987,"Elo_lo":959,"Elo_hi":1016,"n_targets":20,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":978,"Elo_lo":920,"Elo_hi":1035,"n_targets":20,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":956,"Elo_lo":882,"Elo_hi":1033,"n_targets":20,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":933,"Elo_lo":828,"Elo_hi":1036,"n_targets":20,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":898,"Elo_lo":742,"Elo_hi":1043,"n_targets":20,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":833,"Elo_lo":720,"Elo_hi":946,"n_targets":20,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":798,"Elo_lo":657,"Elo_hi":952,"n_targets":20,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":775,"Elo_lo":594,"Elo_hi":942,"n_targets":20,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":749,"Elo_lo":635,"Elo_hi":847,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":484,"Elo_lo":382,"Elo_hi":577,"n_targets":20,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1130,"Elo_lo":976,"Elo_hi":1302,"n_targets":15,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1124,"Elo_lo":920,"Elo_hi":1328,"n_targets":8,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1104,"Elo_lo":951,"Elo_hi":1266,"n_targets":15,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1083,"Elo_lo":946,"Elo_hi":1224,"n_targets":15,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1039,"Elo_lo":895,"Elo_hi":1172,"n_targets":15,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1026,"Elo_lo":936,"Elo_hi":1108,"n_targets":15,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":15,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":988,"Elo_lo":900,"Elo_hi":1074,"n_targets":15,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":987,"Elo_lo":821,"Elo_hi":1158,"n_targets":15,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":987,"Elo_lo":954,"Elo_hi":1015,"n_targets":15,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":979,"Elo_lo":823,"Elo_hi":1128,"n_targets":15,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":947,"Elo_lo":703,"Elo_hi":1195,"n_targets":8,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":942,"Elo_lo":857,"Elo_hi":1025,"n_targets":15,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":866,"Elo_lo":695,"Elo_hi":1047,"n_targets":15,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":828,"Elo_lo":645,"Elo_hi":989,"n_targets":15,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":733,"Elo_lo":425,"Elo_hi":956,"n_targets":15,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":694,"Elo_lo":544,"Elo_hi":821,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":665,"Elo_lo":522,"Elo_hi":806,"n_targets":15,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":551,"Elo_lo":253,"Elo_hi":800,"n_targets":15,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":408,"Elo_lo":223,"Elo_hi":552,"n_targets":15,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1139,"Elo_lo":1078,"Elo_hi":1204,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1111,"Elo_lo":1055,"Elo_hi":1169,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1099,"Elo_lo":1040,"Elo_hi":1164,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1072,"Elo_lo":1030,"Elo_hi":1119,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1067,"Elo_lo":990,"Elo_hi":1140,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1062,"Elo_lo":1004,"Elo_hi":1117,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1057,"Elo_lo":1016,"Elo_hi":1099,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1050,"Elo_lo":984,"Elo_hi":1113,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1049,"Elo_lo":1006,"Elo_hi":1097,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1005,"Elo_lo":925,"Elo_hi":1084,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1001,"Elo_lo":911,"Elo_hi":1081,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":987,"Elo_lo":916,"Elo_hi":1058,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":987,"Elo_lo":948,"Elo_hi":1030,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":986,"Elo_lo":962,"Elo_hi":1006,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":866,"Elo_lo":816,"Elo_hi":917,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":787,"Elo_lo":675,"Elo_hi":880,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":712,"Elo_lo":628,"Elo_hi":783,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":580,"Elo_lo":465,"Elo_hi":688,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":495,"Elo_lo":434,"Elo_hi":543,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1035,"Elo_lo":981,"Elo_hi":1087,"n_targets":40,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1007,"Elo_lo":943,"Elo_hi":1064,"n_targets":40,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":40,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":994,"Elo_lo":969,"Elo_hi":1016,"n_targets":40,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":985,"Elo_lo":922,"Elo_hi":1050,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":980,"Elo_lo":917,"Elo_hi":1046,"n_targets":40,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":973,"Elo_lo":904,"Elo_hi":1038,"n_targets":40,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":965,"Elo_lo":909,"Elo_hi":1025,"n_targets":40,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":958,"Elo_lo":917,"Elo_hi":998,"n_targets":40,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":945,"Elo_lo":891,"Elo_hi":1001,"n_targets":40,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":939,"Elo_lo":869,"Elo_hi":1005,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":929,"Elo_lo":877,"Elo_hi":979,"n_targets":40,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":914,"Elo_lo":860,"Elo_hi":964,"n_targets":40,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":914,"Elo_lo":861,"Elo_hi":963,"n_targets":40,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":886,"Elo_lo":839,"Elo_hi":927,"n_targets":40,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":873,"Elo_lo":817,"Elo_hi":931,"n_targets":40,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":820,"Elo_lo":774,"Elo_hi":864,"n_targets":40,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":746,"Elo_lo":661,"Elo_hi":835,"n_targets":40,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":702,"Elo_lo":630,"Elo_hi":777,"n_targets":40,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":695,"Elo_lo":625,"Elo_hi":770,"n_targets":40,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1062,"Elo_lo":1000,"Elo_hi":1126,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1060,"Elo_lo":1006,"Elo_hi":1112,"n_targets":41,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1018,"Elo_lo":958,"Elo_hi":1080,"n_targets":41,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1015,"Elo_lo":939,"Elo_hi":1094,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":41,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":993,"Elo_lo":925,"Elo_hi":1064,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":990,"Elo_lo":966,"Elo_hi":1013,"n_targets":41,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":982,"Elo_lo":911,"Elo_hi":1053,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":980,"Elo_lo":935,"Elo_hi":1024,"n_targets":41,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":962,"Elo_lo":931,"Elo_hi":992,"n_targets":41,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":960,"Elo_lo":900,"Elo_hi":1026,"n_targets":41,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":944,"Elo_lo":890,"Elo_hi":998,"n_targets":41,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":942,"Elo_lo":883,"Elo_hi":1001,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":937,"Elo_lo":881,"Elo_hi":992,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":921,"Elo_lo":870,"Elo_hi":975,"n_targets":41,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":918,"Elo_lo":861,"Elo_hi":977,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":854,"Elo_lo":807,"Elo_hi":903,"n_targets":41,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":847,"Elo_lo":797,"Elo_hi":898,"n_targets":41,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":778,"Elo_lo":700,"Elo_hi":855,"n_targets":41,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":653,"Elo_lo":588,"Elo_hi":714,"n_targets":41,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1049,"Elo_lo":970,"Elo_hi":1125,"n_targets":40,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1043,"Elo_lo":980,"Elo_hi":1102,"n_targets":40,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1034,"Elo_lo":976,"Elo_hi":1095,"n_targets":40,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1020,"Elo_lo":965,"Elo_hi":1078,"n_targets":40,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1019,"Elo_lo":942,"Elo_hi":1094,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1012,"Elo_lo":933,"Elo_hi":1093,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":40,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":983,"Elo_lo":931,"Elo_hi":1032,"n_targets":40,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":981,"Elo_lo":909,"Elo_hi":1055,"n_targets":40,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":980,"Elo_lo":961,"Elo_hi":1001,"n_targets":40,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":976,"Elo_lo":937,"Elo_hi":1013,"n_targets":40,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":951,"Elo_lo":874,"Elo_hi":1024,"n_targets":40,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":949,"Elo_lo":886,"Elo_hi":1011,"n_targets":40,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":914,"Elo_lo":837,"Elo_hi":989,"n_targets":40,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":911,"Elo_lo":860,"Elo_hi":960,"n_targets":40,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":888,"Elo_lo":846,"Elo_hi":927,"n_targets":40,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":865,"Elo_lo":796,"Elo_hi":930,"n_targets":40,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":860,"Elo_lo":803,"Elo_hi":915,"n_targets":40,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":761,"Elo_lo":674,"Elo_hi":850,"n_targets":40,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":578,"Elo_lo":527,"Elo_hi":629,"n_targets":40,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1106,"Elo_lo":1021,"Elo_hi":1188,"n_targets":32,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1079,"Elo_lo":1001,"Elo_hi":1166,"n_targets":22,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1063,"Elo_lo":1003,"Elo_hi":1121,"n_targets":32,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1052,"Elo_lo":991,"Elo_hi":1109,"n_targets":32,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1043,"Elo_lo":959,"Elo_hi":1128,"n_targets":22,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1040,"Elo_lo":951,"Elo_hi":1119,"n_targets":32,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1027,"Elo_lo":949,"Elo_hi":1103,"n_targets":32,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1013,"Elo_lo":937,"Elo_hi":1103,"n_targets":32,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1011,"Elo_lo":938,"Elo_hi":1079,"n_targets":32,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1002,"Elo_lo":937,"Elo_hi":1064,"n_targets":32,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":32,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":990,"Elo_lo":968,"Elo_hi":1011,"n_targets":32,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":989,"Elo_lo":931,"Elo_hi":1046,"n_targets":32,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":979,"Elo_lo":942,"Elo_hi":1018,"n_targets":32,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":959,"Elo_lo":870,"Elo_hi":1047,"n_targets":32,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":944,"Elo_lo":886,"Elo_hi":997,"n_targets":32,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":917,"Elo_lo":838,"Elo_hi":996,"n_targets":32,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":844,"Elo_lo":776,"Elo_hi":909,"n_targets":32,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":839,"Elo_lo":729,"Elo_hi":941,"n_targets":32,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":552,"Elo_lo":487,"Elo_hi":612,"n_targets":32,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1125,"Elo_lo":1016,"Elo_hi":1229,"n_targets":20,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1078,"Elo_lo":974,"Elo_hi":1181,"n_targets":12,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1068,"Elo_lo":991,"Elo_hi":1149,"n_targets":20,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1066,"Elo_lo":987,"Elo_hi":1143,"n_targets":20,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1055,"Elo_lo":951,"Elo_hi":1177,"n_targets":20,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1036,"Elo_lo":936,"Elo_hi":1137,"n_targets":20,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1025,"Elo_lo":916,"Elo_hi":1130,"n_targets":20,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1009,"Elo_lo":943,"Elo_hi":1073,"n_targets":20,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":20,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":992,"Elo_lo":839,"Elo_hi":1145,"n_targets":12,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":986,"Elo_lo":958,"Elo_hi":1014,"n_targets":20,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":984,"Elo_lo":924,"Elo_hi":1045,"n_targets":20,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":951,"Elo_lo":879,"Elo_hi":1026,"n_targets":20,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":920,"Elo_lo":823,"Elo_hi":1018,"n_targets":20,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":889,"Elo_lo":736,"Elo_hi":1024,"n_targets":20,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":848,"Elo_lo":749,"Elo_hi":950,"n_targets":20,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":797,"Elo_lo":664,"Elo_hi":950,"n_targets":20,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":775,"Elo_lo":602,"Elo_hi":939,"n_targets":20,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":757,"Elo_lo":644,"Elo_hi":853,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":495,"Elo_lo":393,"Elo_hi":589,"n_targets":20,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1130,"Elo_lo":975,"Elo_hi":1302,"n_targets":15,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1126,"Elo_lo":922,"Elo_hi":1330,"n_targets":8,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1104,"Elo_lo":951,"Elo_hi":1267,"n_targets":15,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1083,"Elo_lo":946,"Elo_hi":1226,"n_targets":15,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1039,"Elo_lo":894,"Elo_hi":1174,"n_targets":15,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1026,"Elo_lo":936,"Elo_hi":1109,"n_targets":15,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":15,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":988,"Elo_lo":900,"Elo_hi":1075,"n_targets":15,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":988,"Elo_lo":821,"Elo_hi":1160,"n_targets":15,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":987,"Elo_lo":954,"Elo_hi":1015,"n_targets":15,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":979,"Elo_lo":823,"Elo_hi":1128,"n_targets":15,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":946,"Elo_lo":699,"Elo_hi":1195,"n_targets":8,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":943,"Elo_lo":856,"Elo_hi":1027,"n_targets":15,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":866,"Elo_lo":696,"Elo_hi":1047,"n_targets":15,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":831,"Elo_lo":649,"Elo_hi":991,"n_targets":15,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":734,"Elo_lo":428,"Elo_hi":956,"n_targets":15,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":695,"Elo_lo":546,"Elo_hi":821,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":670,"Elo_lo":529,"Elo_hi":810,"n_targets":15,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":552,"Elo_lo":255,"Elo_hi":801,"n_targets":15,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":410,"Elo_lo":228,"Elo_hi":553,"n_targets":15,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"AUTOGLUON","Elo":1147,"Elo_lo":1083,"Elo_hi":1206,"n_targets":41,"id":"AUTOGLUON","display":"AutoGluon","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"AutoGluon automates tabular learning by training multiple models and combining their predictions in ensembles. Here it runs its native extreme preset within a one-hour budget as a separate AutoML reference.","tuning":null,"category":"AutoML","color":"#111827","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1075,"Elo_lo":1016,"Elo_hi":1136,"n_targets":41,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1072,"Elo_lo":1020,"Elo_hi":1121,"n_targets":41,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1045,"Elo_lo":1004,"Elo_hi":1089,"n_targets":41,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1040,"Elo_lo":987,"Elo_hi":1091,"n_targets":41,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1037,"Elo_lo":963,"Elo_hi":1111,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1032,"Elo_lo":978,"Elo_hi":1084,"n_targets":41,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1025,"Elo_lo":944,"Elo_hi":1106,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1024,"Elo_lo":958,"Elo_hi":1081,"n_targets":41,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":41,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":993,"Elo_lo":954,"Elo_hi":1032,"n_targets":41,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":987,"Elo_lo":910,"Elo_hi":1056,"n_targets":41,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":983,"Elo_lo":901,"Elo_hi":1063,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":982,"Elo_lo":960,"Elo_hi":1005,"n_targets":41,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":980,"Elo_lo":903,"Elo_hi":1051,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":972,"Elo_lo":885,"Elo_hi":1055,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":861,"Elo_lo":814,"Elo_hi":908,"n_targets":41,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":783,"Elo_lo":677,"Elo_hi":883,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":730,"Elo_lo":657,"Elo_hi":803,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":605,"Elo_lo":507,"Elo_hi":702,"n_targets":41,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":530,"Elo_lo":473,"Elo_hi":572,"n_targets":41,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false}],"analysis_views":{"strict":{"reference":[{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1155,"Elo_lo":1098,"Elo_hi":1211,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false,"f1_macro":0.713391503894652,"rmse":1.742894159922996},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1071,"Elo_lo":1016,"Elo_hi":1124,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false,"f1_macro":0.6854581607208678,"rmse":1.8372622165329882},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1063,"Elo_lo":994,"Elo_hi":1126,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.7205882559484305,"rmse":1.955818026331612},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1045,"Elo_lo":988,"Elo_hi":1099,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.7008845111802758,"rmse":2.115544869242223},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1038,"Elo_lo":952,"Elo_hi":1114,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.674900372788093,"rmse":null},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"AUTOGLUON","Elo":1036,"Elo_lo":984,"Elo_hi":1085,"n_targets":42,"id":"AUTOGLUON","display":"AutoGluon","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"AutoGluon automates tabular learning by training multiple models and combining their predictions in ensembles. Here it runs its native extreme preset within a one-hour budget as a separate AutoML reference.","tuning":null,"category":"AutoML","color":"#111827","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6945642015955917,"rmse":2.0725846227782996},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1033,"Elo_lo":979,"Elo_hi":1087,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false,"f1_macro":0.6646341040192109,"rmse":1.8178014682508676},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1031,"Elo_lo":976,"Elo_hi":1085,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true,"f1_macro":0.6939871342451706,"rmse":2.0947178687450423},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1019,"Elo_lo":932,"Elo_hi":1095,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6681470028760327,"rmse":null},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1010,"Elo_lo":963,"Elo_hi":1059,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6812042125308531,"rmse":2.015024780475652},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6658157947879919,"rmse":1.9304502775369867},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":994,"Elo_lo":929,"Elo_hi":1056,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false,"f1_macro":0.6331429574772196,"rmse":1.791143875950348},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":993,"Elo_lo":923,"Elo_hi":1059,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.69916754760155,"rmse":2.1705489086171754},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":973,"Elo_lo":937,"Elo_hi":1011,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6445308419027305,"rmse":1.9776414097758406},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":973,"Elo_lo":945,"Elo_hi":999,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6442367032136745,"rmse":1.998151006713182},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":956,"Elo_lo":900,"Elo_hi":1007,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false,"f1_macro":0.6472922420093705,"rmse":1.902521315844419},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":906,"Elo_lo":851,"Elo_hi":961,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6386446187985723,"rmse":1.8806553715934375},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":867,"Elo_lo":820,"Elo_hi":913,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6349965657930864,"rmse":2.109612748031864},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":865,"Elo_lo":810,"Elo_hi":920,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6290497675098728,"rmse":2.183340886639696},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":744,"Elo_lo":638,"Elo_hi":837,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false,"f1_macro":0.3601253990827079,"rmse":2.167685045473319},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":522,"Elo_lo":457,"Elo_hi":583,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.32326853560954805,"rmse":2.2366634323898165}],"cell_elo":[{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1069,"Elo_lo":1020,"Elo_hi":1117,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1040,"Elo_lo":982,"Elo_hi":1096,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1025,"Elo_lo":977,"Elo_hi":1073,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1019,"Elo_lo":975,"Elo_hi":1064,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1010,"Elo_lo":962,"Elo_hi":1062,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1007,"Elo_lo":937,"Elo_hi":1078,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":1000,"Elo_lo":971,"Elo_hi":1026,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":987,"Elo_lo":948,"Elo_hi":1023,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":970,"Elo_lo":898,"Elo_hi":1041,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":957,"Elo_lo":886,"Elo_hi":1027,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":940,"Elo_lo":883,"Elo_hi":998,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":933,"Elo_lo":883,"Elo_hi":981,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":907,"Elo_lo":856,"Elo_hi":956,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":884,"Elo_lo":819,"Elo_hi":948,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":882,"Elo_lo":825,"Elo_hi":939,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":875,"Elo_lo":791,"Elo_hi":955,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":826,"Elo_lo":778,"Elo_hi":870,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":706,"Elo_lo":630,"Elo_hi":775,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":706,"Elo_lo":629,"Elo_hi":778,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1120,"Elo_lo":1066,"Elo_hi":1173,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1080,"Elo_lo":1024,"Elo_hi":1132,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1079,"Elo_lo":1024,"Elo_hi":1133,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1062,"Elo_lo":1007,"Elo_hi":1116,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1015,"Elo_lo":966,"Elo_hi":1067,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1012,"Elo_lo":938,"Elo_hi":1092,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1008,"Elo_lo":922,"Elo_hi":1094,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1007,"Elo_lo":951,"Elo_hi":1060,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1000,"Elo_lo":952,"Elo_hi":1045,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":986,"Elo_lo":963,"Elo_hi":1011,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":985,"Elo_lo":933,"Elo_hi":1033,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":968,"Elo_lo":914,"Elo_hi":1022,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":963,"Elo_lo":899,"Elo_hi":1027,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":960,"Elo_lo":925,"Elo_hi":995,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":919,"Elo_lo":869,"Elo_hi":970,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":841,"Elo_lo":794,"Elo_hi":883,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":834,"Elo_lo":752,"Elo_hi":910,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":785,"Elo_lo":727,"Elo_hi":841,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":636,"Elo_lo":566,"Elo_hi":701,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1154,"Elo_lo":1092,"Elo_hi":1219,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1112,"Elo_lo":1052,"Elo_hi":1172,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1100,"Elo_lo":1036,"Elo_hi":1166,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1094,"Elo_lo":1034,"Elo_hi":1153,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1090,"Elo_lo":1022,"Elo_hi":1161,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1088,"Elo_lo":1018,"Elo_hi":1154,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1083,"Elo_lo":1018,"Elo_hi":1149,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1066,"Elo_lo":974,"Elo_hi":1152,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1048,"Elo_lo":1002,"Elo_hi":1102,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1046,"Elo_lo":996,"Elo_hi":1098,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1038,"Elo_lo":952,"Elo_hi":1113,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1004,"Elo_lo":932,"Elo_hi":1069,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":988,"Elo_lo":966,"Elo_hi":1009,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":972,"Elo_lo":941,"Elo_hi":1002,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":916,"Elo_lo":860,"Elo_hi":972,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":876,"Elo_lo":824,"Elo_hi":923,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":869,"Elo_lo":806,"Elo_hi":929,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":829,"Elo_lo":740,"Elo_hi":909,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":556,"Elo_lo":487,"Elo_hi":614,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1197,"Elo_lo":1130,"Elo_hi":1269,"n_targets":34,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1152,"Elo_lo":1086,"Elo_hi":1228,"n_targets":34,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1130,"Elo_lo":1043,"Elo_hi":1219,"n_targets":34,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1122,"Elo_lo":1048,"Elo_hi":1196,"n_targets":34,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1118,"Elo_lo":1039,"Elo_hi":1199,"n_targets":34,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1107,"Elo_lo":1037,"Elo_hi":1172,"n_targets":34,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1104,"Elo_lo":1015,"Elo_hi":1196,"n_targets":22,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1098,"Elo_lo":1005,"Elo_hi":1183,"n_targets":22,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1083,"Elo_lo":1015,"Elo_hi":1151,"n_targets":34,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1040,"Elo_lo":949,"Elo_hi":1131,"n_targets":34,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1024,"Elo_lo":959,"Elo_hi":1090,"n_targets":34,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1016,"Elo_lo":942,"Elo_hi":1089,"n_targets":34,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":34,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":993,"Elo_lo":971,"Elo_hi":1014,"n_targets":34,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":987,"Elo_lo":940,"Elo_hi":1033,"n_targets":34,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":966,"Elo_lo":910,"Elo_hi":1016,"n_targets":34,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":919,"Elo_lo":867,"Elo_hi":975,"n_targets":34,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":886,"Elo_lo":777,"Elo_hi":987,"n_targets":34,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":842,"Elo_lo":770,"Elo_hi":914,"n_targets":34,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":498,"Elo_lo":427,"Elo_hi":570,"n_targets":34,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1187,"Elo_lo":1104,"Elo_hi":1276,"n_targets":20,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1135,"Elo_lo":1036,"Elo_hi":1245,"n_targets":20,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1134,"Elo_lo":1021,"Elo_hi":1256,"n_targets":12,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1089,"Elo_lo":1015,"Elo_hi":1168,"n_targets":20,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1088,"Elo_lo":986,"Elo_hi":1204,"n_targets":20,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1063,"Elo_lo":913,"Elo_hi":1226,"n_targets":20,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1049,"Elo_lo":967,"Elo_hi":1138,"n_targets":20,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1047,"Elo_lo":946,"Elo_hi":1145,"n_targets":20,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1030,"Elo_lo":935,"Elo_hi":1134,"n_targets":20,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1030,"Elo_lo":882,"Elo_hi":1180,"n_targets":12,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1004,"Elo_lo":952,"Elo_hi":1056,"n_targets":20,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":20,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":1000,"Elo_lo":967,"Elo_hi":1032,"n_targets":20,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":980,"Elo_lo":914,"Elo_hi":1041,"n_targets":20,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":922,"Elo_lo":811,"Elo_hi":1039,"n_targets":20,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":913,"Elo_lo":856,"Elo_hi":981,"n_targets":20,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":871,"Elo_lo":722,"Elo_hi":1008,"n_targets":20,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":783,"Elo_lo":595,"Elo_hi":948,"n_targets":20,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":757,"Elo_lo":644,"Elo_hi":859,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":448,"Elo_lo":339,"Elo_hi":550,"n_targets":20,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1143,"Elo_lo":974,"Elo_hi":1322,"n_targets":15,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1143,"Elo_lo":987,"Elo_hi":1306,"n_targets":8,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1135,"Elo_lo":998,"Elo_hi":1291,"n_targets":15,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1087,"Elo_lo":955,"Elo_hi":1222,"n_targets":15,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1073,"Elo_lo":936,"Elo_hi":1198,"n_targets":15,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1034,"Elo_lo":964,"Elo_hi":1098,"n_targets":15,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1017,"Elo_lo":811,"Elo_hi":1225,"n_targets":15,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1011,"Elo_lo":857,"Elo_hi":1179,"n_targets":15,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1006,"Elo_lo":934,"Elo_hi":1083,"n_targets":15,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1001,"Elo_lo":849,"Elo_hi":1147,"n_targets":15,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":15,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":983,"Elo_lo":940,"Elo_hi":1018,"n_targets":15,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":965,"Elo_lo":762,"Elo_hi":1171,"n_targets":8,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":955,"Elo_lo":865,"Elo_hi":1043,"n_targets":15,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":859,"Elo_lo":683,"Elo_hi":1014,"n_targets":15,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":751,"Elo_lo":591,"Elo_hi":892,"n_targets":15,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":733,"Elo_lo":431,"Elo_hi":955,"n_targets":15,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":696,"Elo_lo":551,"Elo_hi":816,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":557,"Elo_lo":249,"Elo_hi":806,"n_targets":15,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":403,"Elo_lo":220,"Elo_hi":546,"n_targets":15,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1145,"Elo_lo":1084,"Elo_hi":1210,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1125,"Elo_lo":1070,"Elo_hi":1179,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1123,"Elo_lo":1068,"Elo_hi":1187,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1104,"Elo_lo":1028,"Elo_hi":1177,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1101,"Elo_lo":1054,"Elo_hi":1150,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1081,"Elo_lo":1019,"Elo_hi":1140,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1068,"Elo_lo":1001,"Elo_hi":1136,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1057,"Elo_lo":1012,"Elo_hi":1106,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1050,"Elo_lo":991,"Elo_hi":1106,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1049,"Elo_lo":947,"Elo_hi":1145,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1043,"Elo_lo":1009,"Elo_hi":1079,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1037,"Elo_lo":961,"Elo_hi":1108,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":993,"Elo_lo":951,"Elo_hi":1036,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":990,"Elo_lo":964,"Elo_hi":1012,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":989,"Elo_lo":910,"Elo_hi":1063,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":988,"Elo_lo":904,"Elo_hi":1067,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":881,"Elo_lo":833,"Elo_hi":929,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":623,"Elo_lo":485,"Elo_hi":732,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":494,"Elo_lo":426,"Elo_hi":550,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1049,"Elo_lo":991,"Elo_hi":1108,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1047,"Elo_lo":997,"Elo_hi":1091,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1045,"Elo_lo":997,"Elo_hi":1090,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1011,"Elo_lo":944,"Elo_hi":1078,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":1007,"Elo_lo":986,"Elo_hi":1028,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":996,"Elo_lo":961,"Elo_hi":1026,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":986,"Elo_lo":938,"Elo_hi":1037,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":972,"Elo_lo":902,"Elo_hi":1037,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":966,"Elo_lo":903,"Elo_hi":1026,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":956,"Elo_lo":910,"Elo_hi":997,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":954,"Elo_lo":885,"Elo_hi":1017,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":944,"Elo_lo":897,"Elo_hi":987,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":936,"Elo_lo":889,"Elo_hi":983,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":892,"Elo_lo":848,"Elo_hi":937,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":888,"Elo_lo":835,"Elo_hi":941,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":825,"Elo_lo":781,"Elo_hi":870,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":750,"Elo_lo":666,"Elo_hi":831,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":702,"Elo_lo":626,"Elo_hi":772,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":697,"Elo_lo":622,"Elo_hi":767,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1132,"Elo_lo":1079,"Elo_hi":1182,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1081,"Elo_lo":1028,"Elo_hi":1133,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1081,"Elo_lo":1025,"Elo_hi":1132,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1034,"Elo_lo":947,"Elo_hi":1120,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1025,"Elo_lo":971,"Elo_hi":1078,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1022,"Elo_lo":964,"Elo_hi":1080,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1021,"Elo_lo":944,"Elo_hi":1101,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1006,"Elo_lo":946,"Elo_hi":1068,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":989,"Elo_lo":928,"Elo_hi":1054,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":966,"Elo_hi":1010,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":971,"Elo_lo":916,"Elo_hi":1026,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":970,"Elo_lo":927,"Elo_hi":1011,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":965,"Elo_lo":928,"Elo_hi":1002,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":950,"Elo_lo":897,"Elo_hi":1001,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":927,"Elo_lo":871,"Elo_hi":979,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":835,"Elo_lo":795,"Elo_hi":877,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":824,"Elo_lo":771,"Elo_hi":873,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":763,"Elo_lo":677,"Elo_hi":840,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":625,"Elo_lo":556,"Elo_hi":690,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1155,"Elo_lo":1098,"Elo_hi":1211,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1071,"Elo_lo":1016,"Elo_hi":1124,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1063,"Elo_lo":994,"Elo_hi":1126,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1045,"Elo_lo":988,"Elo_hi":1099,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1038,"Elo_lo":952,"Elo_hi":1114,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"AUTOGLUON","Elo":1036,"Elo_lo":984,"Elo_hi":1085,"n_targets":42,"id":"AUTOGLUON","display":"AutoGluon","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"AutoGluon automates tabular learning by training multiple models and combining their predictions in ensembles. Here it runs its native extreme preset within a one-hour budget as a separate AutoML reference.","tuning":null,"category":"AutoML","color":"#111827","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1033,"Elo_lo":979,"Elo_hi":1087,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1031,"Elo_lo":976,"Elo_hi":1085,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1019,"Elo_lo":932,"Elo_hi":1095,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1010,"Elo_lo":963,"Elo_hi":1059,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":994,"Elo_lo":929,"Elo_hi":1056,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":993,"Elo_lo":923,"Elo_hi":1059,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":973,"Elo_lo":937,"Elo_hi":1011,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":973,"Elo_lo":945,"Elo_hi":999,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":956,"Elo_lo":900,"Elo_hi":1007,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":906,"Elo_lo":851,"Elo_hi":961,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":867,"Elo_lo":820,"Elo_hi":913,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":865,"Elo_lo":810,"Elo_hi":920,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":744,"Elo_lo":638,"Elo_hi":837,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":522,"Elo_lo":457,"Elo_hi":583,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1172,"Elo_lo":1110,"Elo_hi":1234,"n_targets":34,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1123,"Elo_lo":1054,"Elo_hi":1195,"n_targets":34,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1106,"Elo_lo":1008,"Elo_hi":1206,"n_targets":22,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1069,"Elo_lo":991,"Elo_hi":1139,"n_targets":34,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1068,"Elo_lo":998,"Elo_hi":1135,"n_targets":34,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1063,"Elo_lo":980,"Elo_hi":1146,"n_targets":34,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1062,"Elo_lo":960,"Elo_hi":1165,"n_targets":22,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1061,"Elo_lo":997,"Elo_hi":1131,"n_targets":34,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1050,"Elo_lo":965,"Elo_hi":1139,"n_targets":34,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1000,"Elo_lo":958,"Elo_hi":1044,"n_targets":34,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":34,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":994,"Elo_lo":930,"Elo_hi":1058,"n_targets":34,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":984,"Elo_lo":910,"Elo_hi":1062,"n_targets":34,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":977,"Elo_lo":922,"Elo_hi":1033,"n_targets":34,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":977,"Elo_lo":954,"Elo_hi":998,"n_targets":34,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":954,"Elo_lo":879,"Elo_hi":1020,"n_targets":34,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":915,"Elo_lo":853,"Elo_hi":975,"n_targets":34,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":833,"Elo_lo":767,"Elo_hi":895,"n_targets":34,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":817,"Elo_lo":693,"Elo_hi":930,"n_targets":34,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":482,"Elo_lo":414,"Elo_hi":552,"n_targets":34,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1153,"Elo_lo":1066,"Elo_hi":1243,"n_targets":20,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1123,"Elo_lo":1044,"Elo_hi":1218,"n_targets":20,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1094,"Elo_lo":976,"Elo_hi":1215,"n_targets":12,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1066,"Elo_lo":987,"Elo_hi":1149,"n_targets":20,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1065,"Elo_lo":990,"Elo_hi":1144,"n_targets":20,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1045,"Elo_lo":938,"Elo_hi":1154,"n_targets":20,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1029,"Elo_lo":929,"Elo_hi":1123,"n_targets":20,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1020,"Elo_lo":955,"Elo_hi":1086,"n_targets":20,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1012,"Elo_lo":865,"Elo_hi":1166,"n_targets":12,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":20,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":988,"Elo_lo":956,"Elo_hi":1016,"n_targets":20,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":984,"Elo_lo":919,"Elo_hi":1050,"n_targets":20,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":942,"Elo_lo":875,"Elo_hi":1010,"n_targets":20,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":933,"Elo_lo":829,"Elo_hi":1039,"n_targets":20,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":908,"Elo_lo":749,"Elo_hi":1059,"n_targets":20,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":870,"Elo_lo":771,"Elo_hi":968,"n_targets":20,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":806,"Elo_lo":665,"Elo_hi":959,"n_targets":20,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":765,"Elo_lo":578,"Elo_hi":934,"n_targets":20,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":753,"Elo_lo":637,"Elo_hi":849,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":457,"Elo_lo":349,"Elo_hi":557,"n_targets":20,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1112,"Elo_lo":964,"Elo_hi":1277,"n_targets":15,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1112,"Elo_lo":919,"Elo_hi":1300,"n_targets":8,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1100,"Elo_lo":950,"Elo_hi":1258,"n_targets":15,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1063,"Elo_lo":934,"Elo_hi":1193,"n_targets":15,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1044,"Elo_lo":895,"Elo_hi":1177,"n_targets":15,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1015,"Elo_lo":929,"Elo_hi":1094,"n_targets":15,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":15,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":997,"Elo_lo":903,"Elo_hi":1094,"n_targets":15,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":992,"Elo_lo":960,"Elo_hi":1018,"n_targets":15,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":990,"Elo_lo":833,"Elo_hi":1143,"n_targets":15,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":988,"Elo_lo":822,"Elo_hi":1158,"n_targets":15,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":946,"Elo_lo":856,"Elo_hi":1036,"n_targets":15,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":934,"Elo_lo":698,"Elo_hi":1163,"n_targets":8,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":864,"Elo_lo":694,"Elo_hi":1043,"n_targets":15,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":834,"Elo_lo":649,"Elo_hi":997,"n_targets":15,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":739,"Elo_lo":427,"Elo_hi":967,"n_targets":15,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":686,"Elo_lo":542,"Elo_hi":813,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":676,"Elo_lo":529,"Elo_hi":817,"n_targets":15,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":551,"Elo_lo":250,"Elo_hi":798,"n_targets":15,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":405,"Elo_lo":219,"Elo_hi":548,"n_targets":15,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1133,"Elo_lo":1071,"Elo_hi":1198,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1127,"Elo_lo":1070,"Elo_hi":1183,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1116,"Elo_lo":1059,"Elo_hi":1182,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1092,"Elo_lo":1012,"Elo_hi":1170,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1087,"Elo_lo":1025,"Elo_hi":1146,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1083,"Elo_lo":1032,"Elo_hi":1135,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1068,"Elo_lo":999,"Elo_hi":1133,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1063,"Elo_lo":1020,"Elo_hi":1108,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1059,"Elo_lo":1018,"Elo_hi":1103,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1027,"Elo_lo":960,"Elo_hi":1095,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1019,"Elo_lo":942,"Elo_hi":1097,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1004,"Elo_lo":916,"Elo_hi":1083,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":1003,"Elo_lo":957,"Elo_hi":1051,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":978,"Elo_lo":954,"Elo_hi":1003,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":912,"Elo_lo":813,"Elo_hi":1003,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":880,"Elo_lo":834,"Elo_hi":929,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":785,"Elo_lo":714,"Elo_hi":852,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":580,"Elo_lo":443,"Elo_hi":698,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":481,"Elo_lo":412,"Elo_hi":540,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1056,"Elo_lo":1004,"Elo_hi":1105,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1031,"Elo_lo":974,"Elo_hi":1089,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1027,"Elo_lo":977,"Elo_hi":1076,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":967,"Elo_hi":1011,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":982,"Elo_lo":915,"Elo_hi":1050,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":970,"Elo_lo":901,"Elo_hi":1037,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":968,"Elo_lo":914,"Elo_hi":1023,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":965,"Elo_lo":929,"Elo_hi":1001,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":959,"Elo_lo":897,"Elo_hi":1020,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":941,"Elo_lo":869,"Elo_hi":1011,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":929,"Elo_lo":884,"Elo_hi":973,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":921,"Elo_lo":868,"Elo_hi":969,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":908,"Elo_lo":855,"Elo_hi":959,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":871,"Elo_lo":826,"Elo_hi":916,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":865,"Elo_lo":808,"Elo_hi":924,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":793,"Elo_lo":748,"Elo_hi":836,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":736,"Elo_lo":652,"Elo_hi":818,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":687,"Elo_lo":610,"Elo_hi":758,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":685,"Elo_lo":608,"Elo_hi":755,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1140,"Elo_lo":1088,"Elo_hi":1188,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1077,"Elo_lo":1020,"Elo_hi":1131,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1077,"Elo_lo":1022,"Elo_hi":1128,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1023,"Elo_lo":972,"Elo_hi":1078,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1014,"Elo_lo":929,"Elo_hi":1097,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":993,"Elo_lo":921,"Elo_hi":1068,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":986,"Elo_lo":966,"Elo_hi":1006,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":974,"Elo_lo":916,"Elo_hi":1034,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":970,"Elo_lo":924,"Elo_hi":1014,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":962,"Elo_lo":908,"Elo_hi":1020,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":950,"Elo_lo":915,"Elo_hi":987,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":935,"Elo_lo":884,"Elo_hi":990,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":928,"Elo_lo":873,"Elo_hi":982,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":926,"Elo_lo":873,"Elo_hi":979,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":925,"Elo_lo":866,"Elo_hi":980,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":841,"Elo_lo":796,"Elo_hi":887,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":832,"Elo_lo":782,"Elo_hi":880,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":752,"Elo_lo":665,"Elo_hi":832,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":618,"Elo_lo":554,"Elo_hi":682,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1174,"Elo_lo":1115,"Elo_hi":1236,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1082,"Elo_lo":1029,"Elo_hi":1134,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1060,"Elo_lo":992,"Elo_hi":1120,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1044,"Elo_lo":986,"Elo_hi":1106,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1041,"Elo_lo":986,"Elo_hi":1096,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1041,"Elo_lo":955,"Elo_hi":1117,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1028,"Elo_lo":939,"Elo_hi":1109,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1017,"Elo_lo":945,"Elo_hi":1084,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1016,"Elo_lo":965,"Elo_hi":1066,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":983,"Elo_lo":948,"Elo_hi":1019,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":982,"Elo_lo":965,"Elo_hi":1001,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":978,"Elo_lo":910,"Elo_hi":1040,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":973,"Elo_lo":908,"Elo_hi":1038,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":929,"Elo_lo":874,"Elo_hi":980,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":914,"Elo_lo":853,"Elo_hi":977,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":881,"Elo_lo":820,"Elo_hi":936,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":876,"Elo_lo":830,"Elo_hi":920,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":755,"Elo_lo":652,"Elo_hi":842,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":549,"Elo_lo":486,"Elo_hi":604,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1189,"Elo_lo":1132,"Elo_hi":1247,"n_targets":34,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1119,"Elo_lo":1056,"Elo_hi":1186,"n_targets":34,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1085,"Elo_lo":997,"Elo_hi":1171,"n_targets":22,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1068,"Elo_lo":1005,"Elo_hi":1129,"n_targets":34,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1063,"Elo_lo":1006,"Elo_hi":1116,"n_targets":34,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1053,"Elo_lo":958,"Elo_hi":1143,"n_targets":22,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1051,"Elo_lo":971,"Elo_hi":1128,"n_targets":34,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1025,"Elo_lo":945,"Elo_hi":1099,"n_targets":34,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1010,"Elo_lo":939,"Elo_hi":1077,"n_targets":34,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1001,"Elo_lo":966,"Elo_hi":1039,"n_targets":34,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":34,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":998,"Elo_lo":933,"Elo_hi":1056,"n_targets":34,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":992,"Elo_lo":937,"Elo_hi":1043,"n_targets":34,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":968,"Elo_hi":1009,"n_targets":34,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":940,"Elo_lo":851,"Elo_hi":1034,"n_targets":34,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":917,"Elo_lo":861,"Elo_hi":974,"n_targets":34,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":904,"Elo_lo":826,"Elo_hi":974,"n_targets":34,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":845,"Elo_lo":779,"Elo_hi":908,"n_targets":34,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":823,"Elo_lo":708,"Elo_hi":930,"n_targets":34,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":522,"Elo_lo":463,"Elo_hi":579,"n_targets":34,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1157,"Elo_lo":1065,"Elo_hi":1247,"n_targets":20,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1112,"Elo_lo":1027,"Elo_hi":1214,"n_targets":20,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1078,"Elo_lo":976,"Elo_hi":1177,"n_targets":12,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1077,"Elo_lo":997,"Elo_hi":1162,"n_targets":20,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1070,"Elo_lo":989,"Elo_hi":1150,"n_targets":20,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1031,"Elo_lo":932,"Elo_hi":1132,"n_targets":20,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1028,"Elo_lo":916,"Elo_hi":1136,"n_targets":20,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1027,"Elo_lo":961,"Elo_hi":1097,"n_targets":20,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":20,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":992,"Elo_lo":832,"Elo_hi":1149,"n_targets":12,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":987,"Elo_lo":959,"Elo_hi":1016,"n_targets":20,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":978,"Elo_lo":920,"Elo_hi":1035,"n_targets":20,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":956,"Elo_lo":882,"Elo_hi":1033,"n_targets":20,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":933,"Elo_lo":828,"Elo_hi":1036,"n_targets":20,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":898,"Elo_lo":742,"Elo_hi":1043,"n_targets":20,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":833,"Elo_lo":720,"Elo_hi":946,"n_targets":20,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":798,"Elo_lo":657,"Elo_hi":952,"n_targets":20,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":775,"Elo_lo":594,"Elo_hi":942,"n_targets":20,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":749,"Elo_lo":635,"Elo_hi":847,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":484,"Elo_lo":382,"Elo_hi":577,"n_targets":20,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1130,"Elo_lo":976,"Elo_hi":1302,"n_targets":15,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1124,"Elo_lo":920,"Elo_hi":1328,"n_targets":8,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1104,"Elo_lo":951,"Elo_hi":1266,"n_targets":15,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1083,"Elo_lo":946,"Elo_hi":1224,"n_targets":15,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1039,"Elo_lo":895,"Elo_hi":1172,"n_targets":15,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1026,"Elo_lo":936,"Elo_hi":1108,"n_targets":15,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":15,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":988,"Elo_lo":900,"Elo_hi":1074,"n_targets":15,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":987,"Elo_lo":821,"Elo_hi":1158,"n_targets":15,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":987,"Elo_lo":954,"Elo_hi":1015,"n_targets":15,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":979,"Elo_lo":823,"Elo_hi":1128,"n_targets":15,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":947,"Elo_lo":703,"Elo_hi":1195,"n_targets":8,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":942,"Elo_lo":857,"Elo_hi":1025,"n_targets":15,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":866,"Elo_lo":695,"Elo_hi":1047,"n_targets":15,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":828,"Elo_lo":645,"Elo_hi":989,"n_targets":15,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":733,"Elo_lo":425,"Elo_hi":956,"n_targets":15,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":694,"Elo_lo":544,"Elo_hi":821,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":665,"Elo_lo":522,"Elo_hi":806,"n_targets":15,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":551,"Elo_lo":253,"Elo_hi":800,"n_targets":15,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":408,"Elo_lo":223,"Elo_hi":552,"n_targets":15,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1139,"Elo_lo":1078,"Elo_hi":1204,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1111,"Elo_lo":1055,"Elo_hi":1169,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1099,"Elo_lo":1040,"Elo_hi":1164,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1072,"Elo_lo":1030,"Elo_hi":1119,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1067,"Elo_lo":990,"Elo_hi":1140,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1062,"Elo_lo":1004,"Elo_hi":1117,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1057,"Elo_lo":1016,"Elo_hi":1099,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1050,"Elo_lo":984,"Elo_hi":1113,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1049,"Elo_lo":1006,"Elo_hi":1097,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1005,"Elo_lo":925,"Elo_hi":1084,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1001,"Elo_lo":911,"Elo_hi":1081,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":987,"Elo_lo":916,"Elo_hi":1058,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":987,"Elo_lo":948,"Elo_hi":1030,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":986,"Elo_lo":962,"Elo_hi":1006,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":866,"Elo_lo":816,"Elo_hi":917,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":787,"Elo_lo":675,"Elo_hi":880,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":712,"Elo_lo":628,"Elo_hi":783,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":580,"Elo_lo":465,"Elo_hi":688,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":495,"Elo_lo":434,"Elo_hi":543,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1035,"Elo_lo":981,"Elo_hi":1087,"n_targets":40,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1007,"Elo_lo":943,"Elo_hi":1064,"n_targets":40,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":40,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":994,"Elo_lo":969,"Elo_hi":1016,"n_targets":40,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":985,"Elo_lo":922,"Elo_hi":1050,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":980,"Elo_lo":917,"Elo_hi":1046,"n_targets":40,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":973,"Elo_lo":904,"Elo_hi":1038,"n_targets":40,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":965,"Elo_lo":909,"Elo_hi":1025,"n_targets":40,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":958,"Elo_lo":917,"Elo_hi":998,"n_targets":40,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":945,"Elo_lo":891,"Elo_hi":1001,"n_targets":40,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":939,"Elo_lo":869,"Elo_hi":1005,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":929,"Elo_lo":877,"Elo_hi":979,"n_targets":40,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":914,"Elo_lo":860,"Elo_hi":964,"n_targets":40,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":914,"Elo_lo":861,"Elo_hi":963,"n_targets":40,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":886,"Elo_lo":839,"Elo_hi":927,"n_targets":40,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":873,"Elo_lo":817,"Elo_hi":931,"n_targets":40,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":820,"Elo_lo":774,"Elo_hi":864,"n_targets":40,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":746,"Elo_lo":661,"Elo_hi":835,"n_targets":40,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":702,"Elo_lo":630,"Elo_hi":777,"n_targets":40,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":695,"Elo_lo":625,"Elo_hi":770,"n_targets":40,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1062,"Elo_lo":1000,"Elo_hi":1126,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1060,"Elo_lo":1006,"Elo_hi":1112,"n_targets":41,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1018,"Elo_lo":958,"Elo_hi":1080,"n_targets":41,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1015,"Elo_lo":939,"Elo_hi":1094,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":41,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":993,"Elo_lo":925,"Elo_hi":1064,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":990,"Elo_lo":966,"Elo_hi":1013,"n_targets":41,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":982,"Elo_lo":911,"Elo_hi":1053,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":980,"Elo_lo":935,"Elo_hi":1024,"n_targets":41,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":962,"Elo_lo":931,"Elo_hi":992,"n_targets":41,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":960,"Elo_lo":900,"Elo_hi":1026,"n_targets":41,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":944,"Elo_lo":890,"Elo_hi":998,"n_targets":41,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":942,"Elo_lo":883,"Elo_hi":1001,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":937,"Elo_lo":881,"Elo_hi":992,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":921,"Elo_lo":870,"Elo_hi":975,"n_targets":41,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":918,"Elo_lo":861,"Elo_hi":977,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":854,"Elo_lo":807,"Elo_hi":903,"n_targets":41,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":847,"Elo_lo":797,"Elo_hi":898,"n_targets":41,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":778,"Elo_lo":700,"Elo_hi":855,"n_targets":41,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":653,"Elo_lo":588,"Elo_hi":714,"n_targets":41,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1049,"Elo_lo":970,"Elo_hi":1125,"n_targets":40,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1043,"Elo_lo":980,"Elo_hi":1102,"n_targets":40,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1034,"Elo_lo":976,"Elo_hi":1095,"n_targets":40,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1020,"Elo_lo":965,"Elo_hi":1078,"n_targets":40,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1019,"Elo_lo":942,"Elo_hi":1094,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1012,"Elo_lo":933,"Elo_hi":1093,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":40,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":983,"Elo_lo":931,"Elo_hi":1032,"n_targets":40,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":981,"Elo_lo":909,"Elo_hi":1055,"n_targets":40,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":980,"Elo_lo":961,"Elo_hi":1001,"n_targets":40,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":976,"Elo_lo":937,"Elo_hi":1013,"n_targets":40,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":951,"Elo_lo":874,"Elo_hi":1024,"n_targets":40,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":949,"Elo_lo":886,"Elo_hi":1011,"n_targets":40,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":914,"Elo_lo":837,"Elo_hi":989,"n_targets":40,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":911,"Elo_lo":860,"Elo_hi":960,"n_targets":40,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":888,"Elo_lo":846,"Elo_hi":927,"n_targets":40,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":865,"Elo_lo":796,"Elo_hi":930,"n_targets":40,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":860,"Elo_lo":803,"Elo_hi":915,"n_targets":40,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":761,"Elo_lo":674,"Elo_hi":850,"n_targets":40,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":578,"Elo_lo":527,"Elo_hi":629,"n_targets":40,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1106,"Elo_lo":1021,"Elo_hi":1188,"n_targets":32,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1079,"Elo_lo":1001,"Elo_hi":1166,"n_targets":22,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1063,"Elo_lo":1003,"Elo_hi":1121,"n_targets":32,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1052,"Elo_lo":991,"Elo_hi":1109,"n_targets":32,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1043,"Elo_lo":959,"Elo_hi":1128,"n_targets":22,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1040,"Elo_lo":951,"Elo_hi":1119,"n_targets":32,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1027,"Elo_lo":949,"Elo_hi":1103,"n_targets":32,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1013,"Elo_lo":937,"Elo_hi":1103,"n_targets":32,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1011,"Elo_lo":938,"Elo_hi":1079,"n_targets":32,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1002,"Elo_lo":937,"Elo_hi":1064,"n_targets":32,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":32,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":990,"Elo_lo":968,"Elo_hi":1011,"n_targets":32,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":989,"Elo_lo":931,"Elo_hi":1046,"n_targets":32,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":979,"Elo_lo":942,"Elo_hi":1018,"n_targets":32,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":959,"Elo_lo":870,"Elo_hi":1047,"n_targets":32,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":944,"Elo_lo":886,"Elo_hi":997,"n_targets":32,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":917,"Elo_lo":838,"Elo_hi":996,"n_targets":32,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":844,"Elo_lo":776,"Elo_hi":909,"n_targets":32,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":839,"Elo_lo":729,"Elo_hi":941,"n_targets":32,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":552,"Elo_lo":487,"Elo_hi":612,"n_targets":32,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1125,"Elo_lo":1016,"Elo_hi":1229,"n_targets":20,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1078,"Elo_lo":974,"Elo_hi":1181,"n_targets":12,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1068,"Elo_lo":991,"Elo_hi":1149,"n_targets":20,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1066,"Elo_lo":987,"Elo_hi":1143,"n_targets":20,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1055,"Elo_lo":951,"Elo_hi":1177,"n_targets":20,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1036,"Elo_lo":936,"Elo_hi":1137,"n_targets":20,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1025,"Elo_lo":916,"Elo_hi":1130,"n_targets":20,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1009,"Elo_lo":943,"Elo_hi":1073,"n_targets":20,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":20,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":992,"Elo_lo":839,"Elo_hi":1145,"n_targets":12,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":986,"Elo_lo":958,"Elo_hi":1014,"n_targets":20,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":984,"Elo_lo":924,"Elo_hi":1045,"n_targets":20,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":951,"Elo_lo":879,"Elo_hi":1026,"n_targets":20,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":920,"Elo_lo":823,"Elo_hi":1018,"n_targets":20,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":889,"Elo_lo":736,"Elo_hi":1024,"n_targets":20,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":848,"Elo_lo":749,"Elo_hi":950,"n_targets":20,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":797,"Elo_lo":664,"Elo_hi":950,"n_targets":20,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":775,"Elo_lo":602,"Elo_hi":939,"n_targets":20,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":757,"Elo_lo":644,"Elo_hi":853,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":495,"Elo_lo":393,"Elo_hi":589,"n_targets":20,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1130,"Elo_lo":975,"Elo_hi":1302,"n_targets":15,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1126,"Elo_lo":922,"Elo_hi":1330,"n_targets":8,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1104,"Elo_lo":951,"Elo_hi":1267,"n_targets":15,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1083,"Elo_lo":946,"Elo_hi":1226,"n_targets":15,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1039,"Elo_lo":894,"Elo_hi":1174,"n_targets":15,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1026,"Elo_lo":936,"Elo_hi":1109,"n_targets":15,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":15,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":988,"Elo_lo":900,"Elo_hi":1075,"n_targets":15,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":988,"Elo_lo":821,"Elo_hi":1160,"n_targets":15,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":987,"Elo_lo":954,"Elo_hi":1015,"n_targets":15,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":979,"Elo_lo":823,"Elo_hi":1128,"n_targets":15,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":946,"Elo_lo":699,"Elo_hi":1195,"n_targets":8,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":943,"Elo_lo":856,"Elo_hi":1027,"n_targets":15,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":866,"Elo_lo":696,"Elo_hi":1047,"n_targets":15,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":831,"Elo_lo":649,"Elo_hi":991,"n_targets":15,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":734,"Elo_lo":428,"Elo_hi":956,"n_targets":15,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":695,"Elo_lo":546,"Elo_hi":821,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":670,"Elo_lo":529,"Elo_hi":810,"n_targets":15,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":552,"Elo_lo":255,"Elo_hi":801,"n_targets":15,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":410,"Elo_lo":228,"Elo_hi":553,"n_targets":15,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"AUTOGLUON","Elo":1147,"Elo_lo":1083,"Elo_hi":1206,"n_targets":41,"id":"AUTOGLUON","display":"AutoGluon","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"AutoGluon automates tabular learning by training multiple models and combining their predictions in ensembles. Here it runs its native extreme preset within a one-hour budget as a separate AutoML reference.","tuning":null,"category":"AutoML","color":"#111827","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1075,"Elo_lo":1016,"Elo_hi":1136,"n_targets":41,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1072,"Elo_lo":1020,"Elo_hi":1121,"n_targets":41,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1045,"Elo_lo":1004,"Elo_hi":1089,"n_targets":41,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1040,"Elo_lo":987,"Elo_hi":1091,"n_targets":41,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1037,"Elo_lo":963,"Elo_hi":1111,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1032,"Elo_lo":978,"Elo_hi":1084,"n_targets":41,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1025,"Elo_lo":944,"Elo_hi":1106,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1024,"Elo_lo":958,"Elo_hi":1081,"n_targets":41,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":41,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":993,"Elo_lo":954,"Elo_hi":1032,"n_targets":41,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":987,"Elo_lo":910,"Elo_hi":1056,"n_targets":41,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":983,"Elo_lo":901,"Elo_hi":1063,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":982,"Elo_lo":960,"Elo_hi":1005,"n_targets":41,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":980,"Elo_lo":903,"Elo_hi":1051,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":972,"Elo_lo":885,"Elo_hi":1055,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":861,"Elo_lo":814,"Elo_hi":908,"n_targets":41,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":783,"Elo_lo":677,"Elo_hi":883,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":730,"Elo_lo":657,"Elo_hi":803,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":605,"Elo_lo":507,"Elo_hi":702,"n_targets":41,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":530,"Elo_lo":473,"Elo_hi":572,"n_targets":41,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false}]},"adaptive":{"reference":[{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1161,"Elo_lo":1104,"Elo_hi":1216,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false,"f1_macro":0.713391503894652,"rmse":1.648018771835828},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1076,"Elo_lo":1024,"Elo_hi":1130,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false,"f1_macro":0.6854581607208678,"rmse":1.7353954475869495},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1067,"Elo_lo":1004,"Elo_hi":1131,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.7205882559484305,"rmse":1.8467205206793975},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1045,"Elo_lo":989,"Elo_hi":1099,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.7008845111802758,"rmse":2.0009264881764173},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1041,"Elo_lo":965,"Elo_hi":1118,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.674900372788093,"rmse":null},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1039,"Elo_lo":981,"Elo_hi":1095,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false,"f1_macro":0.6646341040192109,"rmse":1.718398306263174},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"AUTOGLUON","Elo":1038,"Elo_lo":988,"Elo_hi":1089,"n_targets":42,"id":"AUTOGLUON","display":"AutoGluon","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"AutoGluon automates tabular learning by training multiple models and combining their predictions in ensembles. Here it runs its native extreme preset within a one-hour budget as a separate AutoML reference.","tuning":null,"category":"AutoML","color":"#111827","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6945642015955917,"rmse":2.0725846227782996},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1032,"Elo_lo":981,"Elo_hi":1085,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true,"f1_macro":0.6939871342451706,"rmse":1.9795885789120025},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1024,"Elo_lo":942,"Elo_hi":1099,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6681470028760327,"rmse":null},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1011,"Elo_lo":963,"Elo_hi":1061,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6812042125308531,"rmse":1.9077731975387953},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6658157947879919,"rmse":1.834015466894516},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":999,"Elo_lo":931,"Elo_hi":1061,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false,"f1_macro":0.6331429574772196,"rmse":1.6968912665990876},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":997,"Elo_lo":927,"Elo_hi":1067,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.69916754760155,"rmse":2.0466117793539307},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":975,"Elo_lo":948,"Elo_hi":1000,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6442367032136745,"rmse":1.8941134993104398},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":972,"Elo_lo":937,"Elo_hi":1009,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6445308419027305,"rmse":1.8793591777811842},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":964,"Elo_lo":909,"Elo_hi":1017,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false,"f1_macro":0.6472922420093705,"rmse":1.7950537882734676},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":910,"Elo_lo":855,"Elo_hi":961,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6386446187985723,"rmse":1.7855562200305728},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":868,"Elo_lo":821,"Elo_hi":912,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6349965657930864,"rmse":2.0070679270455165},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":867,"Elo_lo":811,"Elo_hi":921,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6290497675098728,"rmse":2.183340886639696},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":750,"Elo_lo":642,"Elo_hi":844,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false,"f1_macro":0.3601253990827079,"rmse":2.166151933886909},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":521,"Elo_lo":454,"Elo_hi":580,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.32326853560954805,"rmse":2.1359865882479996}],"cell_elo":[{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1069,"Elo_lo":1020,"Elo_hi":1117,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1040,"Elo_lo":982,"Elo_hi":1096,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1025,"Elo_lo":977,"Elo_hi":1073,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1019,"Elo_lo":975,"Elo_hi":1064,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1010,"Elo_lo":962,"Elo_hi":1062,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1007,"Elo_lo":937,"Elo_hi":1078,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":1000,"Elo_lo":971,"Elo_hi":1026,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":987,"Elo_lo":948,"Elo_hi":1023,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":970,"Elo_lo":898,"Elo_hi":1041,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":957,"Elo_lo":886,"Elo_hi":1027,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":940,"Elo_lo":883,"Elo_hi":998,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":933,"Elo_lo":883,"Elo_hi":981,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":907,"Elo_lo":856,"Elo_hi":956,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":884,"Elo_lo":819,"Elo_hi":948,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":882,"Elo_lo":825,"Elo_hi":939,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":875,"Elo_lo":791,"Elo_hi":955,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":826,"Elo_lo":778,"Elo_hi":870,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":706,"Elo_lo":630,"Elo_hi":775,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":706,"Elo_lo":629,"Elo_hi":778,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1121,"Elo_lo":1067,"Elo_hi":1173,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1081,"Elo_lo":1025,"Elo_hi":1133,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1079,"Elo_lo":1024,"Elo_hi":1133,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1062,"Elo_lo":1006,"Elo_hi":1116,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1016,"Elo_lo":966,"Elo_hi":1067,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1013,"Elo_lo":939,"Elo_hi":1093,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1009,"Elo_lo":923,"Elo_hi":1094,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1008,"Elo_lo":951,"Elo_hi":1061,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1001,"Elo_lo":953,"Elo_hi":1046,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":986,"Elo_lo":962,"Elo_hi":1011,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":985,"Elo_lo":933,"Elo_hi":1033,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":968,"Elo_lo":914,"Elo_hi":1022,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":964,"Elo_lo":899,"Elo_hi":1028,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":960,"Elo_lo":925,"Elo_hi":996,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":919,"Elo_lo":868,"Elo_hi":970,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":859,"Elo_lo":785,"Elo_hi":930,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":840,"Elo_lo":792,"Elo_hi":882,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":784,"Elo_lo":724,"Elo_hi":840,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":634,"Elo_lo":563,"Elo_hi":700,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1159,"Elo_lo":1098,"Elo_hi":1221,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1117,"Elo_lo":1058,"Elo_hi":1178,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1106,"Elo_lo":1038,"Elo_hi":1169,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1096,"Elo_lo":1030,"Elo_hi":1163,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1094,"Elo_lo":1037,"Elo_hi":1155,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1091,"Elo_lo":1024,"Elo_hi":1156,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1087,"Elo_lo":1028,"Elo_hi":1149,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1069,"Elo_lo":983,"Elo_hi":1154,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1048,"Elo_lo":998,"Elo_hi":1098,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1048,"Elo_lo":1003,"Elo_hi":1101,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1042,"Elo_lo":962,"Elo_hi":1117,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1011,"Elo_lo":942,"Elo_hi":1078,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":966,"Elo_hi":1010,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":970,"Elo_lo":939,"Elo_hi":1002,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":919,"Elo_lo":862,"Elo_hi":973,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":884,"Elo_lo":810,"Elo_hi":956,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":875,"Elo_lo":823,"Elo_hi":924,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":870,"Elo_lo":808,"Elo_hi":931,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":547,"Elo_lo":473,"Elo_hi":612,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1209,"Elo_lo":1151,"Elo_hi":1270,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1169,"Elo_lo":1108,"Elo_hi":1237,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1152,"Elo_lo":1075,"Elo_hi":1234,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1132,"Elo_lo":1066,"Elo_hi":1200,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1131,"Elo_lo":1055,"Elo_hi":1208,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1130,"Elo_lo":1062,"Elo_hi":1195,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1119,"Elo_lo":1060,"Elo_hi":1179,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1109,"Elo_lo":1036,"Elo_hi":1183,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1097,"Elo_lo":1036,"Elo_hi":1157,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1068,"Elo_lo":988,"Elo_hi":1145,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1041,"Elo_lo":974,"Elo_hi":1103,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1034,"Elo_lo":976,"Elo_hi":1093,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":995,"Elo_lo":956,"Elo_hi":1035,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":965,"Elo_hi":1011,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":976,"Elo_lo":926,"Elo_hi":1028,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":943,"Elo_lo":844,"Elo_hi":1034,"n_targets":34,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":917,"Elo_lo":869,"Elo_hi":964,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":851,"Elo_lo":781,"Elo_hi":919,"n_targets":34,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":474,"Elo_lo":396,"Elo_hi":543,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1188,"Elo_lo":1137,"Elo_hi":1245,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1161,"Elo_lo":1102,"Elo_hi":1225,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1158,"Elo_lo":1095,"Elo_hi":1226,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1133,"Elo_lo":1043,"Elo_hi":1225,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1110,"Elo_lo":1049,"Elo_hi":1169,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1099,"Elo_lo":1048,"Elo_hi":1153,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1080,"Elo_lo":1003,"Elo_hi":1156,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1078,"Elo_lo":1023,"Elo_hi":1135,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1077,"Elo_lo":1015,"Elo_hi":1135,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1070,"Elo_lo":1003,"Elo_hi":1135,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1009,"Elo_lo":967,"Elo_hi":1050,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1009,"Elo_lo":966,"Elo_hi":1052,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":996,"Elo_lo":929,"Elo_hi":1063,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":994,"Elo_lo":967,"Elo_hi":1017,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":985,"Elo_lo":898,"Elo_hi":1072,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":939,"Elo_lo":816,"Elo_hi":1050,"n_targets":22,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":929,"Elo_lo":883,"Elo_hi":972,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":790,"Elo_lo":675,"Elo_hi":894,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":407,"Elo_lo":315,"Elo_hi":492,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1182,"Elo_lo":1113,"Elo_hi":1250,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1166,"Elo_lo":1097,"Elo_hi":1240,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1153,"Elo_lo":1098,"Elo_hi":1216,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1143,"Elo_lo":1050,"Elo_hi":1237,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1105,"Elo_lo":1045,"Elo_hi":1167,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1098,"Elo_lo":1029,"Elo_hi":1169,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1094,"Elo_lo":1036,"Elo_hi":1153,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1077,"Elo_lo":999,"Elo_hi":1155,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1062,"Elo_lo":993,"Elo_hi":1131,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1032,"Elo_lo":995,"Elo_hi":1070,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1028,"Elo_lo":985,"Elo_hi":1074,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1003,"Elo_lo":919,"Elo_hi":1081,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1000,"Elo_lo":924,"Elo_hi":1070,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":982,"Elo_lo":955,"Elo_hi":1006,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":975,"Elo_lo":873,"Elo_hi":1068,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":959,"Elo_lo":915,"Elo_hi":1003,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":882,"Elo_lo":758,"Elo_hi":997,"n_targets":17,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":770,"Elo_lo":659,"Elo_hi":885,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":415,"Elo_lo":319,"Elo_hi":499,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1146,"Elo_lo":1084,"Elo_hi":1211,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1126,"Elo_lo":1070,"Elo_hi":1182,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1125,"Elo_lo":1069,"Elo_hi":1189,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1112,"Elo_lo":1039,"Elo_hi":1183,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1103,"Elo_lo":1055,"Elo_hi":1154,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1096,"Elo_lo":1005,"Elo_hi":1184,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1082,"Elo_lo":1019,"Elo_hi":1143,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1068,"Elo_lo":999,"Elo_hi":1138,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1058,"Elo_lo":1012,"Elo_hi":1107,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1049,"Elo_lo":989,"Elo_hi":1107,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1042,"Elo_lo":1008,"Elo_hi":1080,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1042,"Elo_lo":965,"Elo_hi":1112,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":994,"Elo_lo":952,"Elo_hi":1037,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":990,"Elo_lo":963,"Elo_hi":1012,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":987,"Elo_lo":901,"Elo_hi":1067,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":977,"Elo_lo":893,"Elo_hi":1055,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":876,"Elo_lo":825,"Elo_hi":925,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":750,"Elo_lo":653,"Elo_hi":837,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":447,"Elo_lo":366,"Elo_hi":515,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1049,"Elo_lo":991,"Elo_hi":1108,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1047,"Elo_lo":997,"Elo_hi":1091,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1045,"Elo_lo":997,"Elo_hi":1090,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1011,"Elo_lo":944,"Elo_hi":1078,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":1007,"Elo_lo":986,"Elo_hi":1028,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":996,"Elo_lo":961,"Elo_hi":1026,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":986,"Elo_lo":938,"Elo_hi":1037,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":972,"Elo_lo":902,"Elo_hi":1037,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":966,"Elo_lo":903,"Elo_hi":1026,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":956,"Elo_lo":910,"Elo_hi":997,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":954,"Elo_lo":885,"Elo_hi":1017,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":944,"Elo_lo":897,"Elo_hi":987,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":936,"Elo_lo":889,"Elo_hi":983,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":892,"Elo_lo":848,"Elo_hi":937,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":888,"Elo_lo":835,"Elo_hi":941,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":825,"Elo_lo":781,"Elo_hi":870,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":750,"Elo_lo":666,"Elo_hi":831,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":702,"Elo_lo":626,"Elo_hi":772,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":697,"Elo_lo":622,"Elo_hi":767,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1132,"Elo_lo":1079,"Elo_hi":1182,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1081,"Elo_lo":1028,"Elo_hi":1133,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1081,"Elo_lo":1025,"Elo_hi":1132,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1034,"Elo_lo":947,"Elo_hi":1120,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1025,"Elo_lo":971,"Elo_hi":1078,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1022,"Elo_lo":964,"Elo_hi":1080,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1021,"Elo_lo":944,"Elo_hi":1101,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1006,"Elo_lo":946,"Elo_hi":1068,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":989,"Elo_lo":928,"Elo_hi":1054,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":966,"Elo_hi":1010,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":971,"Elo_lo":916,"Elo_hi":1026,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":970,"Elo_lo":927,"Elo_hi":1011,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":965,"Elo_lo":928,"Elo_hi":1002,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":950,"Elo_lo":897,"Elo_hi":1001,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":927,"Elo_lo":871,"Elo_hi":979,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":835,"Elo_lo":795,"Elo_hi":877,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":824,"Elo_lo":771,"Elo_hi":873,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":763,"Elo_lo":677,"Elo_hi":840,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":625,"Elo_lo":556,"Elo_hi":690,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1161,"Elo_lo":1104,"Elo_hi":1216,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1076,"Elo_lo":1024,"Elo_hi":1130,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1067,"Elo_lo":1004,"Elo_hi":1131,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1045,"Elo_lo":989,"Elo_hi":1099,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1041,"Elo_lo":965,"Elo_hi":1118,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1039,"Elo_lo":981,"Elo_hi":1095,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"AUTOGLUON","Elo":1038,"Elo_lo":988,"Elo_hi":1089,"n_targets":42,"id":"AUTOGLUON","display":"AutoGluon","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"AutoGluon automates tabular learning by training multiple models and combining their predictions in ensembles. Here it runs its native extreme preset within a one-hour budget as a separate AutoML reference.","tuning":null,"category":"AutoML","color":"#111827","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1032,"Elo_lo":981,"Elo_hi":1085,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1024,"Elo_lo":942,"Elo_hi":1099,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1011,"Elo_lo":963,"Elo_hi":1061,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":999,"Elo_lo":931,"Elo_hi":1061,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":997,"Elo_lo":927,"Elo_hi":1067,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":975,"Elo_lo":948,"Elo_hi":1000,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":972,"Elo_lo":937,"Elo_hi":1009,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":964,"Elo_lo":909,"Elo_hi":1017,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":910,"Elo_lo":855,"Elo_hi":961,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":868,"Elo_lo":821,"Elo_hi":912,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":867,"Elo_lo":811,"Elo_hi":921,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":750,"Elo_lo":642,"Elo_hi":844,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":521,"Elo_lo":454,"Elo_hi":580,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1194,"Elo_lo":1136,"Elo_hi":1255,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1141,"Elo_lo":1081,"Elo_hi":1208,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1135,"Elo_lo":1054,"Elo_hi":1221,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1085,"Elo_lo":1025,"Elo_hi":1145,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1085,"Elo_lo":1024,"Elo_hi":1146,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1083,"Elo_lo":1019,"Elo_hi":1146,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1083,"Elo_lo":1010,"Elo_hi":1162,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1082,"Elo_lo":1001,"Elo_hi":1167,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1078,"Elo_lo":1004,"Elo_hi":1152,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1033,"Elo_lo":965,"Elo_hi":1103,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1005,"Elo_lo":967,"Elo_hi":1045,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":999,"Elo_lo":943,"Elo_hi":1055,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":986,"Elo_lo":940,"Elo_hi":1032,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":972,"Elo_lo":948,"Elo_hi":994,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":955,"Elo_lo":892,"Elo_hi":1016,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":925,"Elo_lo":870,"Elo_hi":980,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":845,"Elo_lo":783,"Elo_hi":904,"n_targets":34,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":832,"Elo_lo":707,"Elo_hi":942,"n_targets":34,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":472,"Elo_lo":409,"Elo_hi":539,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1169,"Elo_lo":1112,"Elo_hi":1232,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1148,"Elo_lo":1093,"Elo_hi":1207,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1138,"Elo_lo":1069,"Elo_hi":1205,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1085,"Elo_lo":1034,"Elo_hi":1139,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1069,"Elo_lo":1009,"Elo_hi":1129,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1064,"Elo_lo":985,"Elo_hi":1140,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1060,"Elo_lo":996,"Elo_hi":1122,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1059,"Elo_lo":1008,"Elo_hi":1111,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1027,"Elo_lo":958,"Elo_hi":1098,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1023,"Elo_lo":983,"Elo_hi":1063,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1017,"Elo_lo":976,"Elo_hi":1060,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1012,"Elo_lo":920,"Elo_hi":1099,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":989,"Elo_lo":901,"Elo_hi":1082,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":978,"Elo_lo":952,"Elo_hi":1004,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":954,"Elo_lo":909,"Elo_hi":999,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":880,"Elo_lo":733,"Elo_hi":1008,"n_targets":21,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":876,"Elo_lo":813,"Elo_hi":942,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":792,"Elo_lo":690,"Elo_hi":888,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":418,"Elo_lo":339,"Elo_hi":499,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1160,"Elo_lo":1095,"Elo_hi":1231,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1147,"Elo_lo":1070,"Elo_hi":1225,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1133,"Elo_lo":1071,"Elo_hi":1200,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1101,"Elo_lo":1042,"Elo_hi":1160,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1073,"Elo_lo":1007,"Elo_hi":1142,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1065,"Elo_lo":1004,"Elo_hi":1128,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1063,"Elo_lo":989,"Elo_hi":1134,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1057,"Elo_lo":964,"Elo_hi":1142,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1056,"Elo_lo":1012,"Elo_hi":1102,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1054,"Elo_lo":961,"Elo_hi":1143,"n_targets":37,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1036,"Elo_lo":992,"Elo_hi":1081,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1024,"Elo_lo":942,"Elo_hi":1101,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":997,"Elo_lo":887,"Elo_hi":1094,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":979,"Elo_lo":931,"Elo_hi":1029,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":976,"Elo_lo":951,"Elo_hi":1000,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":831,"Elo_lo":694,"Elo_hi":959,"n_targets":16,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":799,"Elo_lo":721,"Elo_hi":872,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":770,"Elo_lo":670,"Elo_hi":880,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":405,"Elo_lo":310,"Elo_hi":488,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1135,"Elo_lo":1074,"Elo_hi":1202,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1129,"Elo_lo":1071,"Elo_hi":1187,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1118,"Elo_lo":1060,"Elo_hi":1185,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1101,"Elo_lo":1024,"Elo_hi":1180,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1088,"Elo_lo":1025,"Elo_hi":1148,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1086,"Elo_lo":1033,"Elo_hi":1139,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1069,"Elo_lo":999,"Elo_hi":1134,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1065,"Elo_lo":1021,"Elo_hi":1112,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1061,"Elo_lo":1019,"Elo_hi":1106,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1027,"Elo_lo":958,"Elo_hi":1098,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1027,"Elo_lo":949,"Elo_hi":1103,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1005,"Elo_lo":912,"Elo_hi":1086,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":1005,"Elo_lo":957,"Elo_hi":1054,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":980,"Elo_lo":893,"Elo_hi":1064,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":978,"Elo_lo":953,"Elo_hi":1003,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":871,"Elo_lo":822,"Elo_hi":924,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":765,"Elo_lo":686,"Elo_hi":840,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":602,"Elo_lo":476,"Elo_hi":712,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":439,"Elo_lo":363,"Elo_hi":507,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1056,"Elo_lo":1004,"Elo_hi":1105,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1031,"Elo_lo":974,"Elo_hi":1089,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1027,"Elo_lo":977,"Elo_hi":1076,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":967,"Elo_hi":1011,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":982,"Elo_lo":915,"Elo_hi":1050,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":970,"Elo_lo":901,"Elo_hi":1037,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":968,"Elo_lo":914,"Elo_hi":1023,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":965,"Elo_lo":929,"Elo_hi":1001,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":959,"Elo_lo":897,"Elo_hi":1020,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":941,"Elo_lo":869,"Elo_hi":1011,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":929,"Elo_lo":884,"Elo_hi":973,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":921,"Elo_lo":868,"Elo_hi":969,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":908,"Elo_lo":855,"Elo_hi":959,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":871,"Elo_lo":826,"Elo_hi":916,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":865,"Elo_lo":808,"Elo_hi":924,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":793,"Elo_lo":748,"Elo_hi":836,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":736,"Elo_lo":652,"Elo_hi":818,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":687,"Elo_lo":610,"Elo_hi":758,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":685,"Elo_lo":608,"Elo_hi":755,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1140,"Elo_lo":1088,"Elo_hi":1188,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1077,"Elo_lo":1020,"Elo_hi":1131,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1077,"Elo_lo":1022,"Elo_hi":1128,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1023,"Elo_lo":972,"Elo_hi":1078,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1014,"Elo_lo":929,"Elo_hi":1097,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":993,"Elo_lo":921,"Elo_hi":1068,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":986,"Elo_lo":966,"Elo_hi":1006,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":974,"Elo_lo":916,"Elo_hi":1034,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":970,"Elo_lo":924,"Elo_hi":1014,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":962,"Elo_lo":908,"Elo_hi":1020,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":950,"Elo_lo":915,"Elo_hi":987,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":935,"Elo_lo":884,"Elo_hi":990,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":928,"Elo_lo":873,"Elo_hi":982,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":926,"Elo_lo":873,"Elo_hi":979,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":925,"Elo_lo":866,"Elo_hi":980,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":841,"Elo_lo":796,"Elo_hi":887,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":832,"Elo_lo":782,"Elo_hi":880,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":752,"Elo_lo":665,"Elo_hi":832,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":618,"Elo_lo":554,"Elo_hi":682,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1179,"Elo_lo":1119,"Elo_hi":1238,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1088,"Elo_lo":1037,"Elo_hi":1142,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1064,"Elo_lo":999,"Elo_hi":1129,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1046,"Elo_lo":988,"Elo_hi":1102,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1043,"Elo_lo":965,"Elo_hi":1119,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1042,"Elo_lo":988,"Elo_hi":1097,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1032,"Elo_lo":948,"Elo_hi":1109,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1021,"Elo_lo":950,"Elo_hi":1089,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1017,"Elo_lo":969,"Elo_hi":1068,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":985,"Elo_lo":920,"Elo_hi":1050,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":985,"Elo_lo":966,"Elo_hi":1002,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":982,"Elo_lo":950,"Elo_hi":1017,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":979,"Elo_lo":913,"Elo_hi":1041,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":930,"Elo_lo":877,"Elo_hi":980,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":921,"Elo_lo":859,"Elo_hi":981,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":884,"Elo_lo":825,"Elo_hi":942,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":877,"Elo_lo":830,"Elo_hi":919,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":760,"Elo_lo":660,"Elo_hi":848,"n_targets":42,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":546,"Elo_lo":487,"Elo_hi":603,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1217,"Elo_lo":1162,"Elo_hi":1274,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1146,"Elo_lo":1084,"Elo_hi":1209,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1112,"Elo_lo":1037,"Elo_hi":1188,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1088,"Elo_lo":1036,"Elo_hi":1139,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1087,"Elo_lo":1029,"Elo_hi":1143,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1084,"Elo_lo":1013,"Elo_hi":1151,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1079,"Elo_lo":1000,"Elo_hi":1161,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1045,"Elo_lo":983,"Elo_hi":1105,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1034,"Elo_lo":966,"Elo_hi":1098,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1011,"Elo_lo":979,"Elo_hi":1044,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1005,"Elo_lo":953,"Elo_hi":1057,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":999,"Elo_lo":951,"Elo_hi":1042,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":988,"Elo_lo":966,"Elo_hi":1007,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":969,"Elo_lo":885,"Elo_hi":1054,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":924,"Elo_lo":876,"Elo_hi":975,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":915,"Elo_lo":847,"Elo_hi":977,"n_targets":40,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":859,"Elo_lo":792,"Elo_hi":918,"n_targets":34,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":840,"Elo_lo":727,"Elo_hi":940,"n_targets":34,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":514,"Elo_lo":461,"Elo_hi":567,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1188,"Elo_lo":1133,"Elo_hi":1248,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1141,"Elo_lo":1085,"Elo_hi":1201,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1113,"Elo_lo":1053,"Elo_hi":1176,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1081,"Elo_lo":1029,"Elo_hi":1134,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1067,"Elo_lo":1005,"Elo_hi":1125,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1064,"Elo_lo":1015,"Elo_hi":1115,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1053,"Elo_lo":970,"Elo_hi":1134,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1038,"Elo_lo":976,"Elo_hi":1099,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1021,"Elo_lo":985,"Elo_hi":1056,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1021,"Elo_lo":976,"Elo_hi":1068,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1017,"Elo_lo":927,"Elo_hi":1103,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1017,"Elo_lo":952,"Elo_hi":1086,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":987,"Elo_lo":963,"Elo_hi":1009,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":969,"Elo_lo":874,"Elo_hi":1069,"n_targets":35,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":948,"Elo_lo":905,"Elo_hi":995,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":885,"Elo_lo":745,"Elo_hi":1012,"n_targets":21,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":873,"Elo_lo":798,"Elo_hi":946,"n_targets":34,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":784,"Elo_lo":686,"Elo_hi":874,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":442,"Elo_lo":364,"Elo_hi":519,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1176,"Elo_lo":1111,"Elo_hi":1247,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1128,"Elo_lo":1051,"Elo_hi":1208,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1124,"Elo_lo":1058,"Elo_hi":1194,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1094,"Elo_lo":1033,"Elo_hi":1155,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1056,"Elo_lo":1001,"Elo_hi":1112,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1050,"Elo_lo":984,"Elo_hi":1114,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1048,"Elo_lo":1006,"Elo_hi":1087,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1044,"Elo_lo":950,"Elo_hi":1134,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1041,"Elo_lo":966,"Elo_hi":1116,"n_targets":40,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1028,"Elo_lo":981,"Elo_hi":1077,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1022,"Elo_lo":922,"Elo_hi":1118,"n_targets":30,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1007,"Elo_lo":928,"Elo_hi":1083,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":995,"Elo_lo":891,"Elo_hi":1095,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":980,"Elo_lo":956,"Elo_hi":1002,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":960,"Elo_lo":917,"Elo_hi":1006,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":822,"Elo_lo":685,"Elo_hi":949,"n_targets":16,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":810,"Elo_lo":726,"Elo_hi":890,"n_targets":30,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":767,"Elo_lo":667,"Elo_hi":874,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":408,"Elo_lo":315,"Elo_hi":488,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1141,"Elo_lo":1080,"Elo_hi":1206,"n_targets":43,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1113,"Elo_lo":1055,"Elo_hi":1170,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1101,"Elo_lo":1041,"Elo_hi":1166,"n_targets":43,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1076,"Elo_lo":1001,"Elo_hi":1148,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1074,"Elo_lo":1031,"Elo_hi":1121,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1063,"Elo_lo":1005,"Elo_hi":1119,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1058,"Elo_lo":1017,"Elo_hi":1099,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1051,"Elo_lo":984,"Elo_hi":1113,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1050,"Elo_lo":1007,"Elo_hi":1100,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1012,"Elo_lo":933,"Elo_hi":1089,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1001,"Elo_lo":910,"Elo_hi":1080,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":988,"Elo_lo":948,"Elo_hi":1032,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":987,"Elo_lo":914,"Elo_hi":1059,"n_targets":43,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":985,"Elo_lo":962,"Elo_hi":1006,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":860,"Elo_lo":806,"Elo_hi":912,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":857,"Elo_lo":754,"Elo_hi":948,"n_targets":43,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":694,"Elo_lo":604,"Elo_hi":772,"n_targets":43,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":605,"Elo_lo":499,"Elo_hi":707,"n_targets":43,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":466,"Elo_lo":397,"Elo_hi":522,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1035,"Elo_lo":981,"Elo_hi":1087,"n_targets":40,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1007,"Elo_lo":943,"Elo_hi":1064,"n_targets":40,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":40,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":994,"Elo_lo":969,"Elo_hi":1016,"n_targets":40,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":985,"Elo_lo":922,"Elo_hi":1050,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":980,"Elo_lo":917,"Elo_hi":1046,"n_targets":40,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":973,"Elo_lo":904,"Elo_hi":1038,"n_targets":40,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":965,"Elo_lo":909,"Elo_hi":1025,"n_targets":40,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":958,"Elo_lo":917,"Elo_hi":998,"n_targets":40,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":945,"Elo_lo":891,"Elo_hi":1001,"n_targets":40,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":939,"Elo_lo":869,"Elo_hi":1005,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":929,"Elo_lo":877,"Elo_hi":979,"n_targets":40,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":914,"Elo_lo":860,"Elo_hi":964,"n_targets":40,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":914,"Elo_lo":861,"Elo_hi":963,"n_targets":40,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":886,"Elo_lo":839,"Elo_hi":927,"n_targets":40,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":873,"Elo_lo":817,"Elo_hi":931,"n_targets":40,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":820,"Elo_lo":774,"Elo_hi":864,"n_targets":40,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":746,"Elo_lo":661,"Elo_hi":835,"n_targets":40,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":702,"Elo_lo":630,"Elo_hi":777,"n_targets":40,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":695,"Elo_lo":625,"Elo_hi":770,"n_targets":40,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1062,"Elo_lo":1000,"Elo_hi":1126,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1061,"Elo_lo":1007,"Elo_hi":1112,"n_targets":41,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1018,"Elo_lo":958,"Elo_hi":1081,"n_targets":41,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1015,"Elo_lo":939,"Elo_hi":1094,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":41,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":993,"Elo_lo":925,"Elo_hi":1064,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":990,"Elo_lo":966,"Elo_hi":1013,"n_targets":41,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":988,"Elo_lo":939,"Elo_hi":1036,"n_targets":41,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":982,"Elo_lo":911,"Elo_hi":1053,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":962,"Elo_lo":931,"Elo_hi":992,"n_targets":41,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":960,"Elo_lo":901,"Elo_hi":1026,"n_targets":41,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":944,"Elo_lo":890,"Elo_hi":998,"n_targets":41,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":942,"Elo_lo":883,"Elo_hi":1001,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":937,"Elo_lo":881,"Elo_hi":992,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":921,"Elo_lo":870,"Elo_hi":975,"n_targets":41,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":918,"Elo_lo":861,"Elo_hi":977,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":855,"Elo_lo":807,"Elo_hi":904,"n_targets":41,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":847,"Elo_lo":797,"Elo_hi":898,"n_targets":41,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":778,"Elo_lo":699,"Elo_hi":855,"n_targets":41,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":653,"Elo_lo":589,"Elo_hi":713,"n_targets":41,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1058,"Elo_lo":979,"Elo_hi":1135,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1050,"Elo_lo":988,"Elo_hi":1106,"n_targets":41,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1038,"Elo_lo":980,"Elo_hi":1099,"n_targets":41,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1024,"Elo_lo":970,"Elo_hi":1078,"n_targets":41,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1022,"Elo_lo":944,"Elo_hi":1100,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1016,"Elo_lo":937,"Elo_hi":1097,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":41,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":992,"Elo_lo":945,"Elo_hi":1038,"n_targets":41,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":988,"Elo_lo":918,"Elo_hi":1058,"n_targets":41,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":982,"Elo_lo":962,"Elo_hi":1001,"n_targets":41,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":974,"Elo_lo":935,"Elo_hi":1010,"n_targets":41,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":962,"Elo_lo":880,"Elo_hi":1036,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":956,"Elo_lo":892,"Elo_hi":1020,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":923,"Elo_lo":847,"Elo_hi":996,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":914,"Elo_lo":864,"Elo_hi":961,"n_targets":41,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":888,"Elo_lo":849,"Elo_hi":926,"n_targets":41,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":874,"Elo_lo":806,"Elo_hi":939,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":863,"Elo_lo":808,"Elo_hi":918,"n_targets":40,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":769,"Elo_lo":684,"Elo_hi":854,"n_targets":40,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":577,"Elo_lo":524,"Elo_hi":627,"n_targets":41,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1135,"Elo_lo":1057,"Elo_hi":1206,"n_targets":37,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1098,"Elo_lo":1023,"Elo_hi":1173,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1071,"Elo_lo":1021,"Elo_hi":1120,"n_targets":41,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1071,"Elo_lo":1018,"Elo_hi":1126,"n_targets":40,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1067,"Elo_lo":998,"Elo_hi":1133,"n_targets":41,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1064,"Elo_lo":988,"Elo_hi":1144,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1049,"Elo_lo":962,"Elo_hi":1128,"n_targets":37,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1037,"Elo_lo":976,"Elo_hi":1097,"n_targets":40,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1033,"Elo_lo":965,"Elo_hi":1098,"n_targets":40,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1005,"Elo_lo":952,"Elo_hi":1059,"n_targets":41,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":41,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":992,"Elo_lo":944,"Elo_hi":1040,"n_targets":41,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":988,"Elo_lo":953,"Elo_hi":1024,"n_targets":39,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":988,"Elo_lo":902,"Elo_hi":1073,"n_targets":37,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":985,"Elo_lo":964,"Elo_hi":1006,"n_targets":41,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":947,"Elo_lo":899,"Elo_hi":997,"n_targets":41,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":928,"Elo_lo":854,"Elo_hi":997,"n_targets":37,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":855,"Elo_lo":789,"Elo_hi":919,"n_targets":32,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":855,"Elo_lo":753,"Elo_hi":956,"n_targets":32,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":543,"Elo_lo":482,"Elo_hi":594,"n_targets":41,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1155,"Elo_lo":1090,"Elo_hi":1221,"n_targets":34,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1098,"Elo_lo":1037,"Elo_hi":1162,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1092,"Elo_lo":1016,"Elo_hi":1175,"n_targets":33,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1066,"Elo_lo":1014,"Elo_hi":1123,"n_targets":41,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1066,"Elo_lo":1020,"Elo_hi":1112,"n_targets":40,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1055,"Elo_lo":997,"Elo_hi":1114,"n_targets":41,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1040,"Elo_lo":957,"Elo_hi":1123,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1036,"Elo_lo":973,"Elo_hi":1102,"n_targets":39,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1009,"Elo_lo":974,"Elo_hi":1044,"n_targets":38,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1002,"Elo_lo":960,"Elo_hi":1042,"n_targets":41,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":41,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":996,"Elo_lo":910,"Elo_hi":1077,"n_targets":41,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":994,"Elo_lo":933,"Elo_hi":1055,"n_targets":40,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":975,"Elo_lo":951,"Elo_hi":999,"n_targets":41,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":962,"Elo_lo":867,"Elo_hi":1064,"n_targets":33,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":951,"Elo_lo":908,"Elo_hi":996,"n_targets":41,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":880,"Elo_lo":812,"Elo_hi":945,"n_targets":33,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":874,"Elo_lo":750,"Elo_hi":997,"n_targets":21,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":789,"Elo_lo":695,"Elo_hi":872,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":458,"Elo_lo":376,"Elo_hi":530,"n_targets":41,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1160,"Elo_lo":1085,"Elo_hi":1238,"n_targets":33,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1122,"Elo_lo":1045,"Elo_hi":1202,"n_targets":30,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1110,"Elo_lo":1024,"Elo_hi":1196,"n_targets":28,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1084,"Elo_lo":1024,"Elo_hi":1146,"n_targets":39,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1064,"Elo_lo":1001,"Elo_hi":1128,"n_targets":41,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1037,"Elo_lo":954,"Elo_hi":1117,"n_targets":38,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1036,"Elo_lo":994,"Elo_hi":1077,"n_targets":38,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1028,"Elo_lo":965,"Elo_hi":1093,"n_targets":41,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1028,"Elo_lo":925,"Elo_hi":1125,"n_targets":28,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1024,"Elo_lo":979,"Elo_hi":1071,"n_targets":41,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1018,"Elo_lo":921,"Elo_hi":1125,"n_targets":28,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":41,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":988,"Elo_lo":908,"Elo_hi":1059,"n_targets":40,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":978,"Elo_lo":872,"Elo_hi":1074,"n_targets":41,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":975,"Elo_lo":950,"Elo_hi":999,"n_targets":41,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":966,"Elo_lo":924,"Elo_hi":1008,"n_targets":41,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":815,"Elo_lo":729,"Elo_hi":894,"n_targets":29,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":814,"Elo_lo":696,"Elo_hi":938,"n_targets":16,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":763,"Elo_lo":667,"Elo_hi":865,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":418,"Elo_lo":320,"Elo_hi":494,"n_targets":41,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"AUTOGLUON","Elo":1150,"Elo_lo":1084,"Elo_hi":1211,"n_targets":41,"id":"AUTOGLUON","display":"AutoGluon","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"AutoGluon automates tabular learning by training multiple models and combining their predictions in ensembles. Here it runs its native extreme preset within a one-hour budget as a separate AutoML reference.","tuning":null,"category":"AutoML","color":"#111827","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1093,"Elo_lo":1039,"Elo_hi":1150,"n_targets":41,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1074,"Elo_lo":1022,"Elo_hi":1124,"n_targets":41,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1060,"Elo_lo":992,"Elo_hi":1131,"n_targets":30,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1047,"Elo_lo":1005,"Elo_hi":1092,"n_targets":41,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1041,"Elo_lo":987,"Elo_hi":1093,"n_targets":41,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1033,"Elo_lo":978,"Elo_hi":1085,"n_targets":41,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1027,"Elo_lo":945,"Elo_hi":1108,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1025,"Elo_lo":958,"Elo_hi":1083,"n_targets":41,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1004,"Elo_lo":927,"Elo_hi":1082,"n_targets":30,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":41,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":993,"Elo_lo":954,"Elo_hi":1033,"n_targets":41,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":987,"Elo_lo":908,"Elo_hi":1058,"n_targets":41,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":982,"Elo_lo":960,"Elo_hi":1005,"n_targets":41,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":981,"Elo_lo":901,"Elo_hi":1052,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":975,"Elo_lo":887,"Elo_hi":1057,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":856,"Elo_lo":760,"Elo_hi":951,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":853,"Elo_lo":801,"Elo_hi":902,"n_targets":41,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":710,"Elo_lo":631,"Elo_hi":787,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":624,"Elo_lo":538,"Elo_hi":718,"n_targets":41,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":499,"Elo_lo":432,"Elo_hi":547,"n_targets":41,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false}]},"conditional":{"reference":[{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1173,"Elo_lo":1127,"Elo_hi":1224,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false,"f1_macro":0.7373194688744498,"rmse":1.742894159922996},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1135,"Elo_lo":1046,"Elo_hi":1211,"n_targets":11,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false,"f1_macro":0.6743735459622698,"rmse":0.9646186969717042},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1084,"Elo_lo":1034,"Elo_hi":1132,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false,"f1_macro":0.708422906970535,"rmse":1.8372622165329882},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1066,"Elo_lo":996,"Elo_hi":1131,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.7205882559484305,"rmse":1.955818026331612},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1051,"Elo_lo":967,"Elo_hi":1124,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6975010573849058,"rmse":null},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1047,"Elo_lo":988,"Elo_hi":1102,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.7008845111802758,"rmse":2.115544869242223},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1045,"Elo_lo":994,"Elo_hi":1096,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false,"f1_macro":0.6868807793481313,"rmse":1.8178014682508676},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"AUTOGLUON","Elo":1037,"Elo_lo":984,"Elo_hi":1089,"n_targets":42,"id":"AUTOGLUON","display":"AutoGluon","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"AutoGluon automates tabular learning by training multiple models and combining their predictions in ensembles. Here it runs its native extreme preset within a one-hour budget as a separate AutoML reference.","tuning":null,"category":"AutoML","color":"#111827","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6945642015955917,"rmse":2.0725846227782996},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1033,"Elo_lo":976,"Elo_hi":1089,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true,"f1_macro":0.6939871342451706,"rmse":2.0947178687450423},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1030,"Elo_lo":941,"Elo_hi":1106,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6905148126482918,"rmse":null},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1010,"Elo_lo":962,"Elo_hi":1061,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6812042125308531,"rmse":2.015024780475652},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1005,"Elo_lo":938,"Elo_hi":1068,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false,"f1_macro":0.6543037312012436,"rmse":1.791143875950348},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6658157947879919,"rmse":1.9304502775369867},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":994,"Elo_lo":921,"Elo_hi":1061,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.69916754760155,"rmse":2.1705489086171754},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":986,"Elo_lo":950,"Elo_hi":1026,"n_targets":40,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6751667527954758,"rmse":1.9776414097758406},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":973,"Elo_lo":944,"Elo_hi":999,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6442367032136745,"rmse":1.998151006713182},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":963,"Elo_lo":909,"Elo_hi":1014,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false,"f1_macro":0.6689409220965723,"rmse":1.902521315844419},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":902,"Elo_lo":844,"Elo_hi":960,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6386446187985723,"rmse":1.8806553715934375},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":860,"Elo_lo":811,"Elo_hi":910,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6349965657930864,"rmse":2.109612748031864},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":858,"Elo_lo":800,"Elo_hi":916,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.6290497675098728,"rmse":2.183340886639696},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":482,"Elo_lo":386,"Elo_hi":559,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false,"f1_macro":0.32326853560954805,"rmse":2.2366634323898165}],"cell_elo":[{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1082,"Elo_lo":1040,"Elo_hi":1124,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1041,"Elo_lo":982,"Elo_hi":1098,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1037,"Elo_lo":993,"Elo_hi":1081,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1030,"Elo_lo":992,"Elo_hi":1073,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1021,"Elo_lo":975,"Elo_hi":1070,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1020,"Elo_lo":951,"Elo_hi":1089,"n_targets":28,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":1001,"Elo_lo":970,"Elo_hi":1027,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1000,"Elo_lo":912,"Elo_hi":1090,"n_targets":26,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":987,"Elo_lo":947,"Elo_hi":1023,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":980,"Elo_lo":912,"Elo_hi":1052,"n_targets":28,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":956,"Elo_lo":883,"Elo_hi":1028,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":940,"Elo_lo":894,"Elo_hi":987,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":939,"Elo_lo":879,"Elo_hi":998,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":905,"Elo_lo":853,"Elo_hi":955,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":882,"Elo_lo":815,"Elo_hi":947,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":880,"Elo_lo":821,"Elo_hi":938,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":820,"Elo_lo":768,"Elo_hi":867,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":704,"Elo_lo":621,"Elo_hi":776,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n20","cell_label":"p=2,000, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":704,"Elo_lo":621,"Elo_hi":779,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1136,"Elo_lo":1090,"Elo_hi":1186,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1094,"Elo_lo":1043,"Elo_hi":1144,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1092,"Elo_lo":1039,"Elo_hi":1145,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1063,"Elo_lo":1006,"Elo_hi":1119,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1028,"Elo_lo":956,"Elo_hi":1106,"n_targets":21,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1026,"Elo_lo":977,"Elo_hi":1077,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1026,"Elo_lo":949,"Elo_hi":1107,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1020,"Elo_lo":932,"Elo_hi":1104,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1017,"Elo_lo":963,"Elo_hi":1068,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1000,"Elo_lo":951,"Elo_hi":1047,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":986,"Elo_lo":962,"Elo_hi":1011,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":985,"Elo_lo":933,"Elo_hi":1034,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":967,"Elo_lo":912,"Elo_hi":1023,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":962,"Elo_lo":895,"Elo_hi":1028,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":958,"Elo_lo":922,"Elo_hi":995,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":916,"Elo_lo":865,"Elo_hi":969,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":835,"Elo_lo":784,"Elo_hi":879,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":775,"Elo_lo":715,"Elo_hi":832,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n50","cell_label":"p=2,000, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":624,"Elo_lo":544,"Elo_hi":697,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1171,"Elo_lo":1116,"Elo_hi":1235,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1147,"Elo_lo":1074,"Elo_hi":1224,"n_targets":15,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1128,"Elo_lo":1070,"Elo_hi":1184,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1115,"Elo_lo":1055,"Elo_hi":1179,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1105,"Elo_lo":1038,"Elo_hi":1177,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1098,"Elo_lo":1034,"Elo_hi":1161,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1098,"Elo_lo":1035,"Elo_hi":1160,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1090,"Elo_lo":1019,"Elo_hi":1158,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1079,"Elo_lo":986,"Elo_hi":1163,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1050,"Elo_lo":1002,"Elo_hi":1105,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1050,"Elo_lo":963,"Elo_hi":1126,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1049,"Elo_lo":997,"Elo_hi":1103,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1004,"Elo_lo":929,"Elo_hi":1072,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":988,"Elo_lo":964,"Elo_hi":1010,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":978,"Elo_lo":944,"Elo_hi":1012,"n_targets":40,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":910,"Elo_lo":852,"Elo_hi":970,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":866,"Elo_lo":811,"Elo_hi":918,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":858,"Elo_lo":790,"Elo_hi":924,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n100","cell_label":"p=2,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":515,"Elo_lo":418,"Elo_hi":591,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1243,"Elo_lo":1146,"Elo_hi":1348,"n_targets":13,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1201,"Elo_lo":1132,"Elo_hi":1277,"n_targets":34,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1157,"Elo_lo":1089,"Elo_hi":1234,"n_targets":34,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1133,"Elo_lo":1044,"Elo_hi":1225,"n_targets":34,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1124,"Elo_lo":1047,"Elo_hi":1202,"n_targets":34,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1121,"Elo_lo":1039,"Elo_hi":1205,"n_targets":34,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1109,"Elo_lo":1037,"Elo_hi":1178,"n_targets":34,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1104,"Elo_lo":1013,"Elo_hi":1198,"n_targets":22,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1096,"Elo_lo":1000,"Elo_hi":1184,"n_targets":22,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1085,"Elo_lo":1015,"Elo_hi":1155,"n_targets":34,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1040,"Elo_lo":945,"Elo_hi":1134,"n_targets":34,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1024,"Elo_lo":956,"Elo_hi":1093,"n_targets":34,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1016,"Elo_lo":939,"Elo_hi":1092,"n_targets":34,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":34,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":995,"Elo_lo":944,"Elo_hi":1046,"n_targets":33,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":993,"Elo_lo":970,"Elo_hi":1015,"n_targets":34,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":963,"Elo_lo":905,"Elo_hi":1015,"n_targets":34,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":913,"Elo_lo":858,"Elo_hi":972,"n_targets":34,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":830,"Elo_lo":752,"Elo_hi":906,"n_targets":34,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n200","cell_label":"p=2,000, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":439,"Elo_lo":332,"Elo_hi":528,"n_targets":34,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1191,"Elo_lo":1106,"Elo_hi":1287,"n_targets":20,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1153,"Elo_lo":1011,"Elo_hi":1315,"n_targets":7,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1139,"Elo_lo":1037,"Elo_hi":1258,"n_targets":20,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1132,"Elo_lo":1018,"Elo_hi":1257,"n_targets":12,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1091,"Elo_lo":986,"Elo_hi":1212,"n_targets":20,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1091,"Elo_lo":1016,"Elo_hi":1176,"n_targets":20,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1087,"Elo_lo":934,"Elo_hi":1258,"n_targets":19,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1050,"Elo_lo":966,"Elo_hi":1145,"n_targets":20,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1048,"Elo_lo":944,"Elo_hi":1152,"n_targets":20,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1032,"Elo_lo":933,"Elo_hi":1139,"n_targets":20,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1026,"Elo_lo":870,"Elo_hi":1179,"n_targets":12,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1004,"Elo_lo":950,"Elo_hi":1058,"n_targets":20,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":20,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":1000,"Elo_lo":966,"Elo_hi":1033,"n_targets":20,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":980,"Elo_lo":913,"Elo_hi":1044,"n_targets":20,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":919,"Elo_lo":803,"Elo_hi":1042,"n_targets":20,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":910,"Elo_lo":851,"Elo_hi":980,"n_targets":20,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":868,"Elo_lo":713,"Elo_hi":1010,"n_targets":20,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":739,"Elo_lo":604,"Elo_hi":851,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n500","cell_label":"p=2,000, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":377,"Elo_lo":222,"Elo_hi":510,"n_targets":20,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1145,"Elo_lo":974,"Elo_hi":1335,"n_targets":15,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1143,"Elo_lo":986,"Elo_hi":1315,"n_targets":8,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1137,"Elo_lo":998,"Elo_hi":1301,"n_targets":15,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1088,"Elo_lo":954,"Elo_hi":1232,"n_targets":15,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1079,"Elo_lo":860,"Elo_hi":1200,"n_targets":2,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1075,"Elo_lo":936,"Elo_hi":1207,"n_targets":15,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1054,"Elo_lo":841,"Elo_hi":1277,"n_targets":14,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1035,"Elo_lo":963,"Elo_hi":1101,"n_targets":15,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1011,"Elo_lo":855,"Elo_hi":1186,"n_targets":15,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1006,"Elo_lo":933,"Elo_hi":1087,"n_targets":15,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1001,"Elo_lo":846,"Elo_hi":1155,"n_targets":15,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":15,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":982,"Elo_lo":938,"Elo_hi":1019,"n_targets":15,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":961,"Elo_lo":756,"Elo_hi":1176,"n_targets":8,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":954,"Elo_lo":863,"Elo_hi":1045,"n_targets":15,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":853,"Elo_lo":676,"Elo_hi":1015,"n_targets":15,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":742,"Elo_lo":437,"Elo_hi":963,"n_targets":15,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":738,"Elo_lo":578,"Elo_hi":883,"n_targets":15,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":677,"Elo_lo":518,"Elo_hi":804,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000_n1000","cell_label":"p=2,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":354,"Elo_lo":165,"Elo_hi":508,"n_targets":15,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1203,"Elo_lo":1121,"Elo_hi":1285,"n_targets":31,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1171,"Elo_lo":1069,"Elo_hi":1255,"n_targets":4,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1167,"Elo_lo":1103,"Elo_hi":1230,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1144,"Elo_lo":1071,"Elo_hi":1216,"n_targets":28,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1143,"Elo_lo":1087,"Elo_hi":1206,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1132,"Elo_lo":1072,"Elo_hi":1192,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1107,"Elo_lo":1057,"Elo_hi":1159,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1095,"Elo_lo":1025,"Elo_hi":1165,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1086,"Elo_lo":1020,"Elo_hi":1151,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1069,"Elo_lo":992,"Elo_hi":1143,"n_targets":28,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1063,"Elo_lo":1026,"Elo_hi":1103,"n_targets":41,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1061,"Elo_lo":1013,"Elo_hi":1114,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1053,"Elo_lo":990,"Elo_hi":1115,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1006,"Elo_lo":920,"Elo_hi":1083,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":994,"Elo_lo":907,"Elo_hi":1077,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":993,"Elo_lo":949,"Elo_hi":1040,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":962,"Elo_hi":1013,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":867,"Elo_lo":811,"Elo_hi":921,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_2000","cell_label":"p=2,000, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":374,"Elo_lo":261,"Elo_hi":453,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1060,"Elo_lo":1017,"Elo_hi":1101,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1058,"Elo_lo":1019,"Elo_hi":1099,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1051,"Elo_lo":993,"Elo_hi":1111,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1023,"Elo_lo":960,"Elo_hi":1090,"n_targets":28,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":1007,"Elo_lo":986,"Elo_hi":1029,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":996,"Elo_lo":961,"Elo_hi":1026,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":996,"Elo_lo":951,"Elo_hi":1044,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":982,"Elo_lo":920,"Elo_hi":1048,"n_targets":28,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":965,"Elo_lo":902,"Elo_hi":1026,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":963,"Elo_lo":923,"Elo_hi":1002,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":954,"Elo_lo":884,"Elo_hi":1018,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":953,"Elo_lo":909,"Elo_hi":993,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":939,"Elo_lo":779,"Elo_hi":1089,"n_targets":12,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":935,"Elo_lo":886,"Elo_hi":983,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":891,"Elo_lo":845,"Elo_hi":937,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":887,"Elo_lo":832,"Elo_hi":941,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":822,"Elo_lo":776,"Elo_hi":869,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":703,"Elo_lo":621,"Elo_hi":774,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n20","cell_label":"p=10,000, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":696,"Elo_lo":618,"Elo_hi":768,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1148,"Elo_lo":1102,"Elo_hi":1195,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1095,"Elo_lo":1046,"Elo_hi":1145,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1084,"Elo_lo":1029,"Elo_hi":1136,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1072,"Elo_lo":970,"Elo_hi":1190,"n_targets":11,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1048,"Elo_lo":957,"Elo_hi":1134,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":1036,"Elo_lo":986,"Elo_hi":1088,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1033,"Elo_lo":954,"Elo_hi":1115,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1021,"Elo_lo":961,"Elo_hi":1080,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1018,"Elo_lo":957,"Elo_hi":1079,"n_targets":42,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":990,"Elo_lo":926,"Elo_hi":1056,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":965,"Elo_hi":1010,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":981,"Elo_lo":924,"Elo_hi":1033,"n_targets":42,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":970,"Elo_lo":926,"Elo_hi":1012,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":963,"Elo_lo":925,"Elo_hi":1002,"n_targets":43,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":949,"Elo_lo":894,"Elo_hi":1002,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":924,"Elo_lo":867,"Elo_hi":980,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":828,"Elo_lo":787,"Elo_hi":872,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":815,"Elo_lo":759,"Elo_hi":868,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n50","cell_label":"p=10,000, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":612,"Elo_lo":533,"Elo_hi":685,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1173,"Elo_lo":1127,"Elo_hi":1224,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1135,"Elo_lo":1046,"Elo_hi":1211,"n_targets":11,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1084,"Elo_lo":1034,"Elo_hi":1132,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1066,"Elo_lo":996,"Elo_hi":1131,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1051,"Elo_lo":967,"Elo_hi":1124,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1047,"Elo_lo":988,"Elo_hi":1102,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1045,"Elo_lo":994,"Elo_hi":1096,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"AUTOGLUON","Elo":1037,"Elo_lo":984,"Elo_hi":1089,"n_targets":42,"id":"AUTOGLUON","display":"AutoGluon","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"AutoGluon automates tabular learning by training multiple models and combining their predictions in ensembles. Here it runs its native extreme preset within a one-hour budget as a separate AutoML reference.","tuning":null,"category":"AutoML","color":"#111827","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1033,"Elo_lo":976,"Elo_hi":1089,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1030,"Elo_lo":941,"Elo_hi":1106,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1010,"Elo_lo":962,"Elo_hi":1061,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1005,"Elo_lo":938,"Elo_hi":1068,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":994,"Elo_lo":921,"Elo_hi":1061,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":986,"Elo_lo":950,"Elo_hi":1026,"n_targets":40,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":973,"Elo_lo":944,"Elo_hi":999,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":963,"Elo_lo":909,"Elo_hi":1014,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":902,"Elo_lo":844,"Elo_hi":960,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":860,"Elo_lo":811,"Elo_hi":910,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":858,"Elo_lo":800,"Elo_hi":916,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n100","cell_label":"p=10,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":482,"Elo_lo":386,"Elo_hi":559,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1231,"Elo_lo":1142,"Elo_hi":1321,"n_targets":11,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1175,"Elo_lo":1112,"Elo_hi":1240,"n_targets":34,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1126,"Elo_lo":1056,"Elo_hi":1200,"n_targets":34,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1106,"Elo_lo":1005,"Elo_hi":1207,"n_targets":22,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1070,"Elo_lo":990,"Elo_hi":1143,"n_targets":34,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1070,"Elo_lo":999,"Elo_hi":1140,"n_targets":34,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1065,"Elo_lo":980,"Elo_hi":1150,"n_targets":34,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1063,"Elo_lo":997,"Elo_hi":1133,"n_targets":34,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1060,"Elo_lo":954,"Elo_hi":1163,"n_targets":22,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1052,"Elo_lo":965,"Elo_hi":1142,"n_targets":34,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1012,"Elo_lo":968,"Elo_hi":1058,"n_targets":33,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":34,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":993,"Elo_lo":926,"Elo_hi":1059,"n_targets":34,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":984,"Elo_lo":907,"Elo_hi":1064,"n_targets":34,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":976,"Elo_lo":918,"Elo_hi":1033,"n_targets":34,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":976,"Elo_lo":951,"Elo_hi":998,"n_targets":34,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":952,"Elo_lo":875,"Elo_hi":1021,"n_targets":34,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":910,"Elo_lo":846,"Elo_hi":973,"n_targets":34,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":823,"Elo_lo":752,"Elo_hi":886,"n_targets":34,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n200","cell_label":"p=10,000, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":422,"Elo_lo":321,"Elo_hi":514,"n_targets":34,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1175,"Elo_lo":1032,"Elo_hi":1289,"n_targets":6,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1158,"Elo_lo":1068,"Elo_hi":1256,"n_targets":20,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1128,"Elo_lo":1045,"Elo_hi":1231,"n_targets":20,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1091,"Elo_lo":971,"Elo_hi":1213,"n_targets":12,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1069,"Elo_lo":986,"Elo_hi":1158,"n_targets":20,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1068,"Elo_lo":990,"Elo_hi":1152,"n_targets":20,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1048,"Elo_lo":939,"Elo_hi":1163,"n_targets":20,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1031,"Elo_lo":928,"Elo_hi":1131,"n_targets":20,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1022,"Elo_lo":954,"Elo_hi":1092,"n_targets":20,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1005,"Elo_lo":847,"Elo_hi":1163,"n_targets":12,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":20,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":987,"Elo_lo":954,"Elo_hi":1016,"n_targets":20,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":983,"Elo_lo":915,"Elo_hi":1053,"n_targets":20,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":969,"Elo_lo":807,"Elo_hi":1174,"n_targets":11,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":940,"Elo_lo":870,"Elo_hi":1011,"n_targets":20,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":929,"Elo_lo":821,"Elo_hi":1039,"n_targets":20,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":906,"Elo_lo":741,"Elo_hi":1062,"n_targets":20,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":862,"Elo_lo":756,"Elo_hi":966,"n_targets":20,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":729,"Elo_lo":596,"Elo_hi":838,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n500","cell_label":"p=10,000, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":367,"Elo_lo":203,"Elo_hi":513,"n_targets":20,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1115,"Elo_lo":964,"Elo_hi":1289,"n_targets":15,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1109,"Elo_lo":911,"Elo_hi":1304,"n_targets":8,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1102,"Elo_lo":949,"Elo_hi":1270,"n_targets":15,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1065,"Elo_lo":933,"Elo_hi":1201,"n_targets":15,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1057,"Elo_lo":828,"Elo_hi":1179,"n_targets":2,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1045,"Elo_lo":894,"Elo_hi":1187,"n_targets":15,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1016,"Elo_lo":928,"Elo_hi":1099,"n_targets":15,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":15,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":998,"Elo_lo":799,"Elo_hi":1220,"n_targets":11,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":997,"Elo_lo":902,"Elo_hi":1098,"n_targets":15,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":991,"Elo_lo":958,"Elo_hi":1018,"n_targets":15,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":990,"Elo_lo":829,"Elo_hi":1150,"n_targets":15,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":988,"Elo_lo":818,"Elo_hi":1166,"n_targets":15,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":944,"Elo_lo":854,"Elo_hi":1039,"n_targets":15,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":925,"Elo_lo":681,"Elo_hi":1163,"n_targets":8,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":825,"Elo_lo":638,"Elo_hi":997,"n_targets":15,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":742,"Elo_lo":427,"Elo_hi":976,"n_targets":15,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":669,"Elo_lo":510,"Elo_hi":825,"n_targets":14,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":656,"Elo_lo":511,"Elo_hi":791,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000_n1000","cell_label":"p=10,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":337,"Elo_lo":151,"Elo_hi":495,"n_targets":15,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1205,"Elo_lo":1005,"Elo_hi":1270,"n_targets":3,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1153,"Elo_lo":1090,"Elo_hi":1221,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1139,"Elo_lo":1054,"Elo_hi":1217,"n_targets":23,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1136,"Elo_lo":1078,"Elo_hi":1200,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1134,"Elo_lo":1074,"Elo_hi":1196,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1132,"Elo_lo":1052,"Elo_hi":1209,"n_targets":28,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1091,"Elo_lo":1026,"Elo_hi":1156,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1088,"Elo_lo":1034,"Elo_hi":1143,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1087,"Elo_lo":1042,"Elo_hi":1135,"n_targets":41,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1072,"Elo_lo":999,"Elo_hi":1143,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1063,"Elo_lo":1019,"Elo_hi":1110,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1052,"Elo_lo":971,"Elo_hi":1131,"n_targets":28,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1050,"Elo_lo":980,"Elo_hi":1120,"n_targets":41,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1014,"Elo_lo":922,"Elo_hi":1096,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":1004,"Elo_lo":954,"Elo_hi":1055,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":976,"Elo_lo":949,"Elo_hi":1003,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":865,"Elo_lo":810,"Elo_hi":921,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":781,"Elo_lo":690,"Elo_hi":860,"n_targets":39,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_10000","cell_label":"p=10,000, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":345,"Elo_lo":232,"Elo_hi":433,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1068,"Elo_lo":1024,"Elo_hi":1114,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1039,"Elo_lo":994,"Elo_hi":1085,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1032,"Elo_lo":974,"Elo_hi":1091,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":992,"Elo_lo":929,"Elo_hi":1059,"n_targets":28,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":967,"Elo_hi":1012,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":978,"Elo_lo":927,"Elo_hi":1031,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":970,"Elo_lo":902,"Elo_hi":1037,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":965,"Elo_lo":929,"Elo_hi":1002,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":958,"Elo_lo":895,"Elo_hi":1021,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":949,"Elo_lo":875,"Elo_hi":1021,"n_targets":28,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":937,"Elo_lo":894,"Elo_hi":978,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":924,"Elo_lo":762,"Elo_hi":1072,"n_targets":12,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":919,"Elo_lo":864,"Elo_hi":969,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":913,"Elo_lo":861,"Elo_hi":960,"n_targets":40,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":870,"Elo_lo":824,"Elo_hi":916,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":865,"Elo_lo":808,"Elo_hi":925,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":790,"Elo_lo":745,"Elo_hi":835,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":688,"Elo_lo":606,"Elo_hi":762,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n20","cell_label":"p=25,000, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":683,"Elo_lo":604,"Elo_hi":756,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1157,"Elo_lo":1110,"Elo_hi":1204,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1091,"Elo_lo":1042,"Elo_hi":1142,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1080,"Elo_lo":1022,"Elo_hi":1135,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1059,"Elo_lo":957,"Elo_hi":1180,"n_targets":11,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1027,"Elo_lo":937,"Elo_hi":1113,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1024,"Elo_lo":972,"Elo_hi":1081,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1004,"Elo_lo":930,"Elo_hi":1081,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":986,"Elo_lo":965,"Elo_hi":1006,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":980,"Elo_lo":922,"Elo_hi":1038,"n_targets":41,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":975,"Elo_lo":914,"Elo_hi":1037,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":972,"Elo_lo":926,"Elo_hi":1016,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":955,"Elo_lo":919,"Elo_hi":995,"n_targets":42,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":951,"Elo_lo":891,"Elo_hi":1008,"n_targets":38,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":944,"Elo_lo":893,"Elo_hi":998,"n_targets":42,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":928,"Elo_lo":870,"Elo_hi":984,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":925,"Elo_lo":869,"Elo_hi":980,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":836,"Elo_lo":790,"Elo_hi":884,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":824,"Elo_lo":771,"Elo_hi":875,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n50","cell_label":"p=25,000, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":605,"Elo_lo":528,"Elo_hi":677,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1194,"Elo_lo":1142,"Elo_hi":1252,"n_targets":41,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1137,"Elo_lo":1045,"Elo_hi":1216,"n_targets":11,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1097,"Elo_lo":1049,"Elo_hi":1147,"n_targets":41,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1064,"Elo_lo":994,"Elo_hi":1126,"n_targets":42,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1054,"Elo_lo":966,"Elo_hi":1126,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1047,"Elo_lo":986,"Elo_hi":1110,"n_targets":42,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1043,"Elo_lo":988,"Elo_hi":1101,"n_targets":42,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1041,"Elo_lo":949,"Elo_hi":1119,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1019,"Elo_lo":945,"Elo_hi":1088,"n_targets":42,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1017,"Elo_lo":954,"Elo_hi":1076,"n_targets":37,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1016,"Elo_lo":965,"Elo_hi":1069,"n_targets":42,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1011,"Elo_lo":949,"Elo_hi":1073,"n_targets":38,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1002,"Elo_lo":969,"Elo_hi":1038,"n_targets":39,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":42,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":982,"Elo_lo":963,"Elo_hi":1001,"n_targets":42,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":923,"Elo_lo":865,"Elo_hi":978,"n_targets":42,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":922,"Elo_lo":858,"Elo_hi":986,"n_targets":41,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":873,"Elo_lo":808,"Elo_hi":933,"n_targets":42,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":868,"Elo_lo":817,"Elo_hi":914,"n_targets":42,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n100","cell_label":"p=25,000, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":504,"Elo_lo":412,"Elo_hi":577,"n_targets":42,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1234,"Elo_lo":1146,"Elo_hi":1319,"n_targets":11,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1196,"Elo_lo":1135,"Elo_hi":1258,"n_targets":34,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1124,"Elo_lo":1059,"Elo_hi":1197,"n_targets":34,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1086,"Elo_lo":998,"Elo_hi":1176,"n_targets":22,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1074,"Elo_lo":1009,"Elo_hi":1139,"n_targets":34,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1067,"Elo_lo":1006,"Elo_hi":1124,"n_targets":34,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1065,"Elo_lo":972,"Elo_hi":1162,"n_targets":24,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1055,"Elo_lo":970,"Elo_hi":1137,"n_targets":34,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1054,"Elo_lo":956,"Elo_hi":1145,"n_targets":22,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1047,"Elo_lo":969,"Elo_hi":1118,"n_targets":32,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1012,"Elo_lo":938,"Elo_hi":1083,"n_targets":34,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1009,"Elo_lo":972,"Elo_hi":1047,"n_targets":31,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":34,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":997,"Elo_lo":926,"Elo_hi":1060,"n_targets":34,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":990,"Elo_lo":931,"Elo_hi":1045,"n_targets":34,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":967,"Elo_hi":1010,"n_targets":34,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":961,"Elo_lo":887,"Elo_hi":1034,"n_targets":28,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":910,"Elo_lo":848,"Elo_hi":973,"n_targets":34,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":827,"Elo_lo":751,"Elo_hi":898,"n_targets":34,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n200","cell_label":"p=25,000, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":432,"Elo_lo":327,"Elo_hi":521,"n_targets":34,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1185,"Elo_lo":1037,"Elo_hi":1300,"n_targets":6,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1161,"Elo_lo":1066,"Elo_hi":1262,"n_targets":20,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1116,"Elo_lo":1026,"Elo_hi":1224,"n_targets":20,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1080,"Elo_lo":997,"Elo_hi":1171,"n_targets":20,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1075,"Elo_lo":971,"Elo_hi":1177,"n_targets":12,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1073,"Elo_lo":988,"Elo_hi":1160,"n_targets":20,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1049,"Elo_lo":942,"Elo_hi":1163,"n_targets":19,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1033,"Elo_lo":930,"Elo_hi":1141,"n_targets":20,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1028,"Elo_lo":960,"Elo_hi":1104,"n_targets":20,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":20,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":986,"Elo_lo":957,"Elo_hi":1017,"n_targets":20,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":983,"Elo_lo":924,"Elo_hi":1050,"n_targets":19,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":983,"Elo_lo":818,"Elo_hi":1148,"n_targets":12,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":979,"Elo_lo":820,"Elo_hi":1182,"n_targets":11,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":953,"Elo_lo":876,"Elo_hi":1036,"n_targets":20,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":928,"Elo_lo":819,"Elo_hi":1040,"n_targets":20,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":894,"Elo_lo":727,"Elo_hi":1048,"n_targets":20,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":894,"Elo_lo":776,"Elo_hi":1018,"n_targets":16,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":720,"Elo_lo":584,"Elo_hi":831,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n500","cell_label":"p=25,000, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":389,"Elo_lo":229,"Elo_hi":526,"n_targets":20,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1133,"Elo_lo":975,"Elo_hi":1313,"n_targets":15,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1121,"Elo_lo":910,"Elo_hi":1334,"n_targets":8,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1106,"Elo_lo":950,"Elo_hi":1279,"n_targets":15,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1084,"Elo_lo":945,"Elo_hi":1235,"n_targets":15,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1059,"Elo_lo":830,"Elo_hi":1184,"n_targets":2,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1040,"Elo_lo":894,"Elo_hi":1182,"n_targets":15,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1027,"Elo_lo":934,"Elo_hi":1113,"n_targets":15,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":15,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":999,"Elo_lo":800,"Elo_hi":1226,"n_targets":11,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":988,"Elo_lo":899,"Elo_hi":1079,"n_targets":15,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":987,"Elo_lo":952,"Elo_hi":1016,"n_targets":15,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":986,"Elo_lo":816,"Elo_hi":1168,"n_targets":15,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":978,"Elo_lo":819,"Elo_hi":1136,"n_targets":15,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":940,"Elo_lo":854,"Elo_hi":1027,"n_targets":15,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":938,"Elo_lo":679,"Elo_hi":1193,"n_targets":8,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":818,"Elo_lo":635,"Elo_hi":987,"n_targets":15,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":733,"Elo_lo":426,"Elo_hi":964,"n_targets":15,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":679,"Elo_lo":523,"Elo_hi":838,"n_targets":13,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":664,"Elo_lo":516,"Elo_hi":799,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000_n1000","cell_label":"p=25,000, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":339,"Elo_lo":151,"Elo_hi":497,"n_targets":15,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1188,"Elo_lo":995,"Elo_hi":1244,"n_targets":3,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1159,"Elo_lo":1098,"Elo_hi":1225,"n_targets":42,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1118,"Elo_lo":1059,"Elo_hi":1182,"n_targets":42,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1117,"Elo_lo":1058,"Elo_hi":1179,"n_targets":43,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1106,"Elo_lo":1033,"Elo_hi":1179,"n_targets":28,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1094,"Elo_lo":992,"Elo_hi":1191,"n_targets":15,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1088,"Elo_lo":1045,"Elo_hi":1132,"n_targets":40,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1077,"Elo_lo":1032,"Elo_hi":1126,"n_targets":43,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1066,"Elo_lo":1005,"Elo_hi":1126,"n_targets":43,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1054,"Elo_lo":983,"Elo_hi":1121,"n_targets":43,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1053,"Elo_lo":1007,"Elo_hi":1105,"n_targets":43,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1037,"Elo_lo":954,"Elo_hi":1117,"n_targets":28,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1024,"Elo_lo":948,"Elo_hi":1097,"n_targets":39,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1012,"Elo_lo":918,"Elo_hi":1094,"n_targets":43,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":43,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":986,"Elo_lo":943,"Elo_hi":1032,"n_targets":43,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":985,"Elo_lo":960,"Elo_hi":1007,"n_targets":43,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":851,"Elo_lo":793,"Elo_hi":908,"n_targets":43,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":798,"Elo_lo":690,"Elo_hi":887,"n_targets":26,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_25000","cell_label":"p=25,000, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":348,"Elo_lo":239,"Elo_hi":426,"n_targets":43,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1071,"Elo_lo":1025,"Elo_hi":1118,"n_targets":34,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1069,"Elo_lo":1022,"Elo_hi":1115,"n_targets":31,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1036,"Elo_lo":979,"Elo_hi":1091,"n_targets":40,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":40,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"XT","Elo":996,"Elo_lo":973,"Elo_hi":1020,"n_targets":39,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":987,"Elo_lo":926,"Elo_hi":1052,"n_targets":28,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":983,"Elo_lo":933,"Elo_hi":1041,"n_targets":35,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":979,"Elo_lo":913,"Elo_hi":1046,"n_targets":40,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":968,"Elo_lo":927,"Elo_hi":1009,"n_targets":36,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":960,"Elo_lo":903,"Elo_hi":1021,"n_targets":40,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":954,"Elo_lo":910,"Elo_hi":1000,"n_targets":37,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":938,"Elo_lo":867,"Elo_hi":1008,"n_targets":28,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":929,"Elo_lo":780,"Elo_hi":1078,"n_targets":12,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":920,"Elo_lo":864,"Elo_hi":970,"n_targets":37,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":918,"Elo_lo":861,"Elo_hi":967,"n_targets":39,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":880,"Elo_lo":832,"Elo_hi":924,"n_targets":40,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":872,"Elo_lo":810,"Elo_hi":934,"n_targets":40,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":808,"Elo_lo":757,"Elo_hi":857,"n_targets":40,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":692,"Elo_lo":611,"Elo_hi":774,"n_targets":40,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n20","cell_label":"p=all, n=20","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":686,"Elo_lo":606,"Elo_hi":769,"n_targets":40,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1151,"Elo_lo":1108,"Elo_hi":1195,"n_targets":34,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1092,"Elo_lo":1038,"Elo_hi":1142,"n_targets":31,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1069,"Elo_lo":1009,"Elo_hi":1126,"n_targets":40,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1065,"Elo_lo":957,"Elo_hi":1171,"n_targets":11,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1017,"Elo_lo":933,"Elo_hi":1101,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1013,"Elo_lo":950,"Elo_hi":1075,"n_targets":41,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":40,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":996,"Elo_lo":939,"Elo_hi":1057,"n_targets":34,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":994,"Elo_lo":918,"Elo_hi":1069,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"XT","Elo":990,"Elo_lo":965,"Elo_hi":1015,"n_targets":40,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":981,"Elo_lo":931,"Elo_hi":1029,"n_targets":40,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":966,"Elo_lo":907,"Elo_hi":1028,"n_targets":33,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":962,"Elo_lo":925,"Elo_hi":996,"n_targets":37,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":960,"Elo_lo":899,"Elo_hi":1023,"n_targets":40,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":954,"Elo_lo":895,"Elo_hi":1013,"n_targets":38,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":942,"Elo_lo":884,"Elo_hi":999,"n_targets":41,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":915,"Elo_lo":859,"Elo_hi":976,"n_targets":40,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":832,"Elo_lo":784,"Elo_hi":877,"n_targets":41,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":826,"Elo_lo":767,"Elo_hi":883,"n_targets":40,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n50","cell_label":"p=all, n=50","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":619,"Elo_lo":538,"Elo_hi":692,"n_targets":41,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1174,"Elo_lo":1124,"Elo_hi":1228,"n_targets":32,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1127,"Elo_lo":1029,"Elo_hi":1217,"n_targets":11,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1081,"Elo_lo":1030,"Elo_hi":1129,"n_targets":29,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1051,"Elo_lo":985,"Elo_hi":1115,"n_targets":40,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1039,"Elo_lo":976,"Elo_hi":1106,"n_targets":40,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1027,"Elo_lo":970,"Elo_hi":1087,"n_targets":40,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1027,"Elo_lo":945,"Elo_hi":1104,"n_targets":29,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1022,"Elo_lo":933,"Elo_hi":1103,"n_targets":29,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1020,"Elo_lo":946,"Elo_hi":1092,"n_targets":28,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":40,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":998,"Elo_lo":964,"Elo_hi":1035,"n_targets":37,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":993,"Elo_lo":918,"Elo_hi":1074,"n_targets":39,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":993,"Elo_lo":942,"Elo_hi":1045,"n_targets":39,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":980,"Elo_lo":920,"Elo_hi":1039,"n_targets":37,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"XT","Elo":978,"Elo_lo":957,"Elo_hi":1001,"n_targets":40,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":941,"Elo_lo":876,"Elo_hi":1006,"n_targets":30,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":900,"Elo_lo":840,"Elo_hi":954,"n_targets":40,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":875,"Elo_lo":829,"Elo_hi":920,"n_targets":40,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":849,"Elo_lo":785,"Elo_hi":911,"n_targets":39,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n100","cell_label":"p=all, n=100","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":511,"Elo_lo":421,"Elo_hi":587,"n_targets":40,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1228,"Elo_lo":1140,"Elo_hi":1313,"n_targets":11,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1190,"Elo_lo":1128,"Elo_hi":1252,"n_targets":28,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1129,"Elo_lo":1058,"Elo_hi":1213,"n_targets":25,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1080,"Elo_lo":996,"Elo_hi":1171,"n_targets":22,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1069,"Elo_lo":1003,"Elo_hi":1135,"n_targets":32,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1064,"Elo_lo":972,"Elo_hi":1168,"n_targets":24,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1058,"Elo_lo":992,"Elo_hi":1121,"n_targets":32,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1052,"Elo_lo":974,"Elo_hi":1128,"n_targets":30,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"LR","Elo":1046,"Elo_lo":948,"Elo_hi":1130,"n_targets":32,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1040,"Elo_lo":947,"Elo_hi":1133,"n_targets":22,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1014,"Elo_lo":935,"Elo_hi":1089,"n_targets":32,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1001,"Elo_lo":964,"Elo_hi":1038,"n_targets":28,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1001,"Elo_lo":928,"Elo_hi":1069,"n_targets":32,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":32,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":989,"Elo_lo":924,"Elo_hi":1052,"n_targets":32,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"XT","Elo":989,"Elo_lo":964,"Elo_hi":1012,"n_targets":32,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":987,"Elo_lo":903,"Elo_hi":1071,"n_targets":26,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":936,"Elo_lo":871,"Elo_hi":997,"n_targets":32,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":821,"Elo_lo":744,"Elo_hi":897,"n_targets":32,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n200","cell_label":"p=all, n=200","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":452,"Elo_lo":354,"Elo_hi":543,"n_targets":32,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1180,"Elo_lo":1035,"Elo_hi":1299,"n_targets":6,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1167,"Elo_lo":1068,"Elo_hi":1270,"n_targets":19,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1153,"Elo_lo":1050,"Elo_hi":1272,"n_targets":17,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1074,"Elo_lo":991,"Elo_hi":1164,"n_targets":20,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1074,"Elo_lo":967,"Elo_hi":1185,"n_targets":12,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1071,"Elo_lo":986,"Elo_hi":1158,"n_targets":20,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1047,"Elo_lo":942,"Elo_hi":1165,"n_targets":19,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1040,"Elo_lo":933,"Elo_hi":1151,"n_targets":20,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1011,"Elo_lo":941,"Elo_hi":1082,"n_targets":20,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":20,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"XT","Elo":985,"Elo_lo":954,"Elo_hi":1015,"n_targets":20,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":983,"Elo_lo":920,"Elo_hi":1051,"n_targets":20,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":980,"Elo_lo":814,"Elo_hi":1147,"n_targets":12,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":977,"Elo_lo":818,"Elo_hi":1183,"n_targets":11,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":947,"Elo_lo":871,"Elo_hi":1029,"n_targets":20,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":913,"Elo_lo":810,"Elo_hi":1020,"n_targets":20,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":912,"Elo_lo":802,"Elo_hi":1027,"n_targets":16,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"LR","Elo":881,"Elo_lo":721,"Elo_hi":1030,"n_targets":20,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":723,"Elo_lo":586,"Elo_hi":834,"n_targets":20,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n500","cell_label":"p=all, n=500","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":387,"Elo_lo":227,"Elo_hi":524,"n_targets":20,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1132,"Elo_lo":975,"Elo_hi":1314,"n_targets":15,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1123,"Elo_lo":914,"Elo_hi":1336,"n_targets":8,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1107,"Elo_lo":951,"Elo_hi":1279,"n_targets":15,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1085,"Elo_lo":945,"Elo_hi":1236,"n_targets":15,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1060,"Elo_lo":832,"Elo_hi":1185,"n_targets":2,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1040,"Elo_lo":893,"Elo_hi":1185,"n_targets":15,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1027,"Elo_lo":934,"Elo_hi":1114,"n_targets":15,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":15,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":999,"Elo_lo":799,"Elo_hi":1227,"n_targets":11,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":988,"Elo_lo":898,"Elo_hi":1079,"n_targets":15,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":987,"Elo_lo":816,"Elo_hi":1169,"n_targets":15,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"XT","Elo":987,"Elo_lo":952,"Elo_hi":1016,"n_targets":15,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":979,"Elo_lo":820,"Elo_hi":1137,"n_targets":15,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":941,"Elo_lo":854,"Elo_hi":1029,"n_targets":15,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":936,"Elo_lo":679,"Elo_hi":1194,"n_targets":8,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":821,"Elo_lo":639,"Elo_hi":990,"n_targets":15,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"LR","Elo":735,"Elo_lo":428,"Elo_hi":965,"n_targets":15,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":685,"Elo_lo":528,"Elo_hi":839,"n_targets":13,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":665,"Elo_lo":519,"Elo_hi":799,"n_targets":15,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full_n1000","cell_label":"p=all, n=1,000","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":341,"Elo_lo":155,"Elo_hi":498,"n_targets":15,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"AUTOGLUON","Elo":1176,"Elo_lo":1113,"Elo_hi":1236,"n_targets":40,"id":"AUTOGLUON","display":"AutoGluon","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"AutoGluon automates tabular learning by training multiple models and combining their predictions in ensembles. Here it runs its native extreme preset within a one-hour budget as a separate AutoML reference.","tuning":null,"category":"AutoML","color":"#111827","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABICL","Elo":1169,"Elo_lo":998,"Elo_hi":1237,"n_targets":3,"id":"TABICL","display":"TabICL v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/soda-inria/tabicl","description":"TabICL v2 is a transformer pretrained on varied synthetic tables for classification and regression. It uses labelled training rows as context when predicting new rows, with attention designed to handle larger tables without requiring equally long tables throughout pretraining.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2.5","Elo":1133,"Elo_lo":1067,"Elo_hi":1201,"n_targets":33,"id":"REALTABPFN-V2.5","display":"RealTabPFN 2.5","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 2.5, by Prior Labs, uses a pretrained transformer to make predictions from labelled examples, extending the v2 family to wider tables. Its default classifier includes pretraining on real datasets, while its default regressor is pretrained on synthetic data only.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABFM","Elo":1116,"Elo_lo":1015,"Elo_hi":1223,"n_targets":13,"id":"TABFM","display":"TabFM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/google-research/tabfm","description":"TabFM is Google's pretrained model for classification and regression on tables containing numerical and categorical columns. It predicts from labelled training rows supplied as context, using sampled groups of rows and features when needed, rather than updating its model weights for each new dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABM","Elo":1109,"Elo_lo":1047,"Elo_hi":1171,"n_targets":39,"id":"TABM","display":"TabM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/yandex-research/tabm","description":"TabM, by Yandex Research, trains a group of multilayer perceptrons together for tabular classification or regression, with most of their weights shared. Each input produces several predictions that are combined, giving the model an ensemble structure without storing and training a full set of separate networks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-V3","Elo":1108,"Elo_lo":1037,"Elo_hi":1187,"n_targets":30,"id":"TABPFN-V3","display":"TabPFN 3","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN 3, by Prior Labs, is a transformer pretrained on synthetic prediction tasks. For classification or regression on a new table, it uses the labelled training rows as examples to guide its predictions rather than learning the task from scratch.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":10000,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE-5K-NE3","Elo":1086,"Elo_lo":1011,"Elo_hi":1166,"n_targets":27,"id":"TABPFN-WIDE-5K-NE3","display":"TabPFN Wide 5k (ne3)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This TabPFN-Wide configuration uses the wide-v2-5k checkpoint, pretrained for classification with many features and few samples. It averages three inference variants while keeping the checkpoint's original feature grouping.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"CAT","Elo":1084,"Elo_lo":1035,"Elo_hi":1130,"n_targets":36,"id":"CAT","display":"CatBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/catboost/catboost","description":"CatBoost builds an ensemble of decision trees through gradient boosting, adding trees that improve the current predictions. It supports numerical and categorical features for classification and regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABDPT","Elo":1080,"Elo_lo":1023,"Elo_hi":1135,"n_targets":41,"id":"TABDPT","display":"TabDPT","tuned_from":null,"classification_only":false,"source_url":"https://github.com/layer6ai-labs/TabDPT-inference","description":"TabDPT is a tabular foundation model by Layer6 AI pretrained on real datasets using self-supervised prediction tasks. For a new classification or regression problem, it retrieves relevant labelled rows and uses them as context to predict new rows without updating its weights on that dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":2500,"training_data_overlap":true},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"GBM","Elo":1051,"Elo_lo":1005,"Elo_hi":1102,"n_targets":41,"id":"GBM","display":"LightGBM","tuned_from":null,"classification_only":false,"source_url":"https://github.com/lightgbm-org/LightGBM","description":"LightGBM is a gradient boosting method that builds decision trees using binned feature values. It grows each tree by splitting the leaf with the largest reduction in the training objective.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"NN_TORCH","Elo":1044,"Elo_lo":985,"Elo_hi":1102,"n_targets":41,"id":"NN_TORCH","display":"MLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"This multilayer perceptron uses AutoGluon's PyTorch implementation for tabular classification and regression. It passes the input features through fully connected hidden layers with nonlinear activations to learn the prediction function.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"REALMLP","Elo":1038,"Elo_lo":970,"Elo_hi":1100,"n_targets":40,"id":"REALMLP","display":"RealMLP","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dholzmueller/pytabkit","description":"RealMLP is a multilayer perceptron trained on each dataset for classification or regression. It combines numerical preprocessing with a carefully chosen network and training setup, using defaults selected across a broad collection of tabular tasks.","tuning":null,"category":"Deep Learning","color":"#dc2626","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"REALTABPFN-V2","Elo":1021,"Elo_lo":941,"Elo_hi":1097,"n_targets":37,"id":"REALTABPFN-V2","display":"RealTabPFN v2","tuned_from":null,"classification_only":false,"source_url":"https://github.com/PriorLabs/TabPFN","description":"TabPFN v2, by Prior Labs, is a pretrained transformer for classification and regression on tables. It takes labelled training rows as context and predicts new rows without training a separate neural network from scratch for each dataset.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"TABPFN-WIDE","Elo":1021,"Elo_lo":928,"Elo_hi":1110,"n_targets":27,"id":"TABPFN-WIDE","display":"TabPFN Wide (8k)","tuned_from":null,"classification_only":true,"source_url":"REMOVED","description":"TabPFN-Wide continues pre-training TabPFN 2 on synthetic tables with many features and few samples, a setting common in omics. This classification model uses the wide-v2-8k checkpoint to work with wider inputs than the original model.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"RF","Elo":1000,"Elo_lo":1000,"Elo_hi":1000,"n_targets":41,"id":"RF","display":"Random Forest","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Random Forest grows decision trees on bootstrap samples of the training data, considering a random subset of features at each split. It averages their class probabilities for classification or their numerical predictions for regression.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"LR","Elo":998,"Elo_lo":914,"Elo_hi":1071,"n_targets":41,"id":"LR","display":"Logistic Regression","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Logistic regression learns a weighted combination of the input features and turns it into class probabilities. For continuous targets, this entry uses ridge regression, a linear model with a penalty that keeps its coefficients from becoming too large.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"XGB","Elo":991,"Elo_lo":947,"Elo_hi":1038,"n_targets":41,"id":"XGB","display":"XGBoost","tuned_from":null,"classification_only":false,"source_url":"https://github.com/dmlc/xgboost","description":"XGBoost combines decision trees through gradient boosting, adding trees to improve the ensemble's predictions. Its training objective includes penalties on tree complexity, alongside the loss for classification or regression.","tuning":null,"category":"Gradient Boosting","color":"#d97706","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"XT","Elo":981,"Elo_lo":955,"Elo_hi":1006,"n_targets":41,"id":"XT","display":"Extra Trees","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"Extra Trees averages the predictions of many decision trees. Each tree considers randomly chosen features and split thresholds, adding more randomness than a conventional random forest when learning a classification or regression task.","tuning":null,"category":"Tree-based","color":"#059669","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"KNN","Elo":844,"Elo_lo":786,"Elo_hi":899,"n_targets":40,"id":"KNN","display":"KNN","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"KNN predicts from the training examples closest to a new sample in feature space. The benchmark uses equally weighted neighbors, taking their class votes for classification or averaging their target values for regression.","tuning":null,"category":"Traditional ML","color":"#2563eb","regular_max_features":null,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"MITRA","Elo":806,"Elo_lo":699,"Elo_hi":896,"n_targets":25,"id":"MITRA","display":"MITRA","tuned_from":null,"classification_only":false,"source_url":"https://github.com/autogluon/autogluon","description":"MITRA is a pretrained model for tabular classification and regression that learns from a mixture of synthetic datasets generated by different rules. It uses labelled examples as context for new predictions, and AutoGluon's implementation also adapts its pretrained weights to the target dataset through fine-tuning.","tuning":null,"category":"Tabular Foundation","color":"#7c3aed","regular_max_features":500,"training_data_overlap":false},{"cell":"cap_full","cell_label":"p=all, n=all","domain":"all","metric":"f1_macro","model_id":"DUMMY","Elo":352,"Elo_lo":250,"Elo_hi":436,"n_targets":41,"id":"DUMMY","display":"Constant","tuned_from":null,"classification_only":false,"source_url":"https://github.com/scikit-learn/scikit-learn","description":"The constant baseline ignores the input features. It returns the training set's class proportions for classification or its mean target value for regression, showing what can be predicted without using the measurements.","tuning":null,"category":"Baseline","color":"#9ca3af","regular_max_features":null,"training_data_overlap":false}]}}}
