BEHAVIOR-1K 2026 Challenge Leaderboard

BEHAVIOR-1K 2026 Challenge Leaderboard

Shows the public rankings and a submission portal for the BEHAVIOR-1K 2026 Challenge, plus the archived 2025 board. For challenge entrants, organizers and anyone following the results.

Made by behavior-1k 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: This Space shows the public rankings for the BEHAVIOR-1K 2026 Challenge and takes entries through a portal. An older board for the 2025 challenge is archived beside it. Organizers run the evaluation by hand on the official OmniGibson setup, then add the verified results to files in the repository.

Who it suits: It suits teams entering the challenge, the organizers who maintain the results, and researchers who follow the rankings.

Industry | Who uses it | How they use it | Example
Benchmarking | Challenge entrants | Read the rankings and send an entry through the submission portal | Look at the 2026 board, then open the submission page
Benchmarking | Challenge organizers | Keep the public results files current after manual evaluation | Append one verified score line to data/results.jsonl
Research | Researchers | Compare entries on the 2026 board with the archived 2025 board | Read both boards side by side

If you do not follow this challenge, you can skip it.

Where to use it:

• A web browser, on the Space's page on Hugging Face
• A local copy of the files, run on Python 3.11 under Windows, macOS or Linux
• The repository's data folder, where organizers edit the results files

Use case | Who | What they get back
Check the standings | Entrants and followers | The public 2026 leaderboard and the archived 2025 one
Submit an entry | Entrants | A response saved as a JSON file in a private Hugging Face dataset
Update results | Organizers | Public files that hold submission metadata and verified scores

How to set it up:

• A web browser, which is all the online page needs
• Python 3.11, for a local copy
• The Python packages listed in requirements.txt
• To host your own portal, the SUBMISSION_DATASET_REPO setting and an HF_WRITE_TOKEN secret with write access to that private dataset

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 5.35.0, then the packages from requirements.txt.
5. Start the app with python app.py.

pip install -U huggingface_hub
hf download behavior-1k/2026-challenge-leaderboard --repo-type space --local-dir 2026-challenge-leaderboard
cd 2026-challenge-leaderboard
pip install gradio==5.35.0
pip install -r requirements.txt
python app.py

How to use it: Visitors open the page to read the leaderboard and use its submit page to enter. Organizers keep the results files up to date by appending JSON lines, and a script can rebuild the self-reported aggregates from the private submission files.

Tool | What it does
data/submissions.jsonl | Holds public submission metadata
data/results.jsonl | Holds aggregate verified scores
data/published_self_reported_results.jsonl | Holds fallback scores the organizers approved for the public
scripts/extract_self_reported_scores.py | Writes self-reported aggregate scores to a private path you choose, outside the repository

HF_TOKEN=<private-dataset-read-token> python scripts/extract_self_reported_scores.py --output /private/path/self_reported_results.jsonl

Example: A team opens the Space, reads the 2026 board and sends an entry through the submit page. The portal saves the response as a JSON file under 2026/submissions/ in the private dataset. Organizers then evaluate it by hand before any verified score is added.

For newcomers and experts: Open the Space and look at the leaderboard first. That shows who has entered and how the scores compare, and you can ignore the repository files unless you run the challenge.

Each score is the mean across 100 challenge tasks, and any task left unsubmitted adds zero. Verified results come from manual runs on the official OmniGibson setup. The public app loads only the self-reported scores that organizers cleared. The detailed per-case data and the script that reproduces runs are kept private.

An entrant submits through the portal
The response is saved as JSON in a private dataset
Organizers evaluate it on the OmniGibson setup
Verified rows are appended to the data files
The public app reads them for the leaderboard

Good to know: Results appear only after the organizers add them, because evaluation is manual. Public files hold only visible leaderboard fields and technical row IDs.

• Not for: posting participant identities, private files, credentials or case-level metrics, and nothing goes public without the organizers' approval
• License: Apache 2.0
• Cost to run: a write token as the HF_WRITE_TOKEN secret for the private dataset, and a read token as HF_TOKEN for the extraction script
• Data it reaches: the public data folder in the repository and a private Hugging Face dataset that holds portal responses

On Hugging Face, 13 November 2026

• Likes: 30
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/behavior-1k/2026-challenge-leaderboard
• Files: https://huggingface.co/spaces/behavior-1k/2026-challenge-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.