TabArena

TabArena

Ranks machine learning methods for tabular data with Elo ratings in a Gradio leaderboard, and serves the numbers to agents as JSON and MCP tools. For researchers and practitioners choosing tabular ML methods.

Made by TabArena on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: TabArena is the leaderboard front end for TabArena, a benchmark that ranks machine learning methods for tabular data with Elo ratings. It runs on Hugging Face as a Gradio app and also covers a second benchmark called BeyondArena. The published numbers are plain CSV files in the repository.

Who it suits: It suits researchers and data scientists who need to compare methods for tabular data, and developers who want to read the rankings from code or an AI agent.

Industry | Who uses it | How they use it | Example
Machine learning research | Researchers | Compare methods on shared tabular benchmarks | Open the leaderboard and sort by Elo
Data science | Practitioners | Choose a method for a tabular problem | Look at which entrants rank highest
AI agents | Developers | Fetch rankings through JSON or MCP tools | Call get_tabarena_leaderboard from an agent

If your data is images or text rather than tables, you can skip it.

Where to use it:

• A web browser, on the Space's page on Hugging Face
• An MCP-capable agent, using the four MCP tools the Space serves
• A local copy of the files, run with Python under Windows, macOS or Linux

Use case | Who | What they get back
Read the leaderboard | Researchers | Elo-ranked tables for tabular methods
Query from an agent | Developers | Rankings through list_leaderboards and three more tools
Update the results | Maintainers | A refreshed data folder after the website script runs

How to set it up:

• A web browser, which is all the online page needs
• Python, for a local copy
• The Python packages listed in requirements.txt

1. Open the Space's page on Hugging Face to use it online. Nothing needs installing.
2. For a local copy, install the Hugging Face hub package and download the Space's files.
3. Move into the downloaded folder.
4. Install Gradio 6.22.0, then the packages from requirements.txt.
5. Start the app with python main.py.

pip install -U huggingface_hub
hf download TabArena/leaderboard --repo-type space --local-dir leaderboard
cd leaderboard
pip install gradio==6.22.0
pip install -r requirements.txt
python main.py

For development work on the leaderboard code, the README gives two install commands.

pip install -e ".[dev]"
uv pip install -r pyproject.toml

How to use it: Open the page and read the leaderboards. Programs can read the same data through four tools defined in api.py.

Tool | What it does
list_leaderboards | Lists the leaderboards that are available
get_tabarena_leaderboard | Returns the TabArena leaderboard
get_beyondarena_leaderboard | Returns the BeyondArena leaderboard
get_pareto_frontier | Returns the Pareto frontier

Example: An agent calls list_leaderboards to see what exists, then calls get_tabarena_leaderboard. It gets back the ranked table without opening the web page.

For newcomers and experts: Open the Space page and look at the main table first. Elo here is a rating from pairwise comparisons, so a higher number means a method beat others more often.

The entrants setting decides who competed. Individual models only is the default, and the other keys add open-source local systems, systems with an LLM in the loop and systems behind a closed API. Choosing a pool does not just hide rows. Elo depends on the entrants, so a different pool gives a fresh evaluation and different numbers. The leaderboard pages are packed in data/pages.sqlite.

Benchmark runs are turned into website artifacts
The artifacts are copied into the data folder
The app reads the data and the packed pages
The page and the four tools serve the rankings

Good to know: Refreshing the results is a maintainer job. Test the leaderboard locally after the change.

• Refresh steps: run the website-artifacts script from the TabArena codebase, remove the old entrants data, the dataset count file and pages.sqlite, then copy the new website data into the data folder
• License: Apache 2.0
• Cost to run: none named, and the online copy runs on basic CPU hardware
• Data it reaches: the CSV files, the packed pages and the data_beyondarena folder in its own repository

On Hugging Face, 13 November 2026

• Likes: 222
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/TabArena/leaderboard
• Files: https://huggingface.co/spaces/TabArena/leaderboard/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.