TabArena is a living benchmarking system that makes benchmarking tabular machine learning models a reliable experience. TabArena implements best practices to ensure methods are represented at their peak potential, including cross-validated ensembles, strong hyperparameter search spaces contributed by the method authors, early stopping, model refitting, parallel bagging, memory usage estimation, and more.
TabArena currently consists of:
- 51 manually curated tabular datasets representing real-world tabular data tasks.
- 9 to 30 evaluated splits per dataset.
- 27+ tabular machine learning methods, including 10+ tabular foundation models.
- More than 50 million trained models across the benchmark, with all validation and test predictions cached to enable tuning and post-hoc ensembling analysis.
- A live TabArena leaderboard showcasing the results.
Tip
The fastest way to try TabArena end-to-end:
pip install uv
git clone https://github.com/autogluon/tabarena.git && cd tabarena
uv sync --extra benchmark
uv run python examples/benchmarking/run_quickstart_tabarena.pyFor other install paths (eval-only, editable AutoGluon, dependency), see Installation below.
We share more details on various use cases of TabArena in our examples:
- 📊 Benchmarking Predictive Machine Learning Models: please refer to examples/benchmarking.
- 🚀 Using SOTA Tabular Models Benchmarked by TabArena: please refer to examples/running_tabarena_models.
- 🗃️ Analysing Metadata and Meta-Learning: please refer to examples/meta.
- 📈 Generating Plots and Leaderboards: please refer to examples/plots_and_leaderboards.
- 🔁 Reproducibility: we share instructions for reproducibility in examples.
Please refer to our dataset curation repository to learn more about or contributed data!
TabArena code is currently being polished. Detailed Documentation for TabArena will be available soon.
Important
Requires Python 3.11–3.13 and uv.
Pick the install path that matches what you want to do:
📊 Evaluation only — leaderboards & metrics, no model fitting
git clone https://github.com/autogluon/tabarena.git
cd tabarena
uv sync🚀 Benchmark — core set of models for benchmarking
Installs the core models used for standard benchmarking: tabpfn, tabicl, ebm, search_spaces, realmlp, tabdpt, tabm.
git clone https://github.com/autogluon/tabarena.git
cd tabarena
uv sync --extra benchmark➕ Benchmark + Extended — core models plus the extended model set
The
extendedextra is experimental and may fail to resolve or install due to incompatible version requirements across model dependencies. Use it only if you specifically need every model in a single environment; otherwise preferbenchmarkorbenchmarkplus one specific model.
Layers the extended model set (modernnca, xrfm, sap-rpt-oss, ...) on top of the core benchmark set.
git clone https://github.com/autogluon/tabarena.git
cd tabarena
uv sync --extra benchmark --extra extendedTo install only one extended model on top of benchmark (recommended over extended when you only need a single extra model), pass its extra by name — for example, just xrfm:
git clone https://github.com/autogluon/tabarena.git
cd tabarena
uv sync --extra benchmark --extra xrfm🛠️ Developer — editable AutoGluon + editable TabArena
Create a virtual environment:
uv venv --seed --python 3.12 ~/.venvs/tabarena
source ~/.venvs/tabarena/bin/activateInstall editable AutoGluon and TabArena:
git clone https://github.com/autogluon/autogluon.git
./autogluon/full_install.sh
git clone https://github.com/autogluon/tabarena.git
uv pip install --prerelease=allow -e "./tabarena/tabarena[benchmark]"In PyCharm, mark
tabarena/and eachautogluon/src/subdirectory as Sources Root so imports resolve.
📦 Use TabArena as a dependency
Add the following to your project's dependencies:
"tabarena @ git+https://github.com/autogluon/tabarena.git#subdirectory=tabarena"TabArena caches predictions, results, and leaderboards as downloadable artifacts so you can reproduce or extend any analysis without re-running the benchmark.
Artifact tiers, sizes, and examples
Artifacts download to
~/.cache/tabarena/by default. Override the location with theTABARENA_CACHEenvironment variable.Raw data is ~100 GB per method type. Point
TABARENA_CACHEat a large disk before downloading it.
| Tier | Contents | Size / method | Example |
|---|---|---|---|
| Raw data | Per-child test predictions, full metadata, system info | ~100 GB | inspect_raw_data.py |
| Processed data | Minimal data for HPO simulation, portfolios, leaderboards | ~10 GB | inspect_processed_data.py |
| Results | Per-config / HPO DataFrames (test error, val error, train time, inference time) | <1 MB | run_generate_main_leaderboard.py |
| Leaderboards | Aggregated ELO, win-rate, average rank, improvability | <1 MB | — |
| Figures & Plots | Generated from results and leaderboards | — | — |
[!TIP] If you use TabArena in a scientific publication, please cite our paper.
TabArena: A Living Benchmark for Machine Learning on Tabular Data Nick Erickson, Lennart Purucker, Andrej Tschalzev, David Holzmüller, Prateek Mutalik Desai, David Salinas, Frank Hutter NeurIPS 2025, Datasets and Benchmarks Track
📄 arXiv · 🎤 NeurIPS poster & video
BibTeX
The entry uses
year=2026because NeurIPS'25 proceedings are published in 2026.
@article{erickson2026tabarena,
title = {TabArena: A Living Benchmark for Machine Learning on Tabular Data},
author = {Erickson, Nick and Purucker, Lennart and Tschalzev, Andrej and Holzm{\"u}ller, David and Desai, Prateek and Salinas, David and Hutter, Frank},
journal = {Advances in Neural Information Processing Systems},
volume = {38},
year = {2026}
}TabArena was built upon and now replaces TabRepo. To see details about TabRepo, the portfolio simulation repository, refer to tabrepo.md.