Thanks for submitting your work! This guide covers everything you need to know.
- Fork this repository
- Create a new directory under
submissions/with a unique, lowercase, hyphenated ID:submissions/my-method-name/ - Add a
submission.yamlfile in your directory. Copy the template fromsubmissions/.example/submission.yamland fill in your details. - Open a Pull Request against the
mainbranch.
- Automated checks will run on your PR to validate formatting and required fields.
- If checks fail, review the error messages and push fixes to your PR branch.
- A maintainer reviews passing PRs on a roughly weekly cadence.
- Once merged, your entry appears on the leaderboard.
To update your submission (e.g. improved score, new version):
- Add a new entry to the
versionslist in your existingsubmission.yaml - Open a PR with the changes
Do not create a new directory for updates — keep everything in your original submission directory.
-
name— unique display name for your method -
authors— at least one author with aname -
description— brief explanation of your approach -
code_url— link to a public repository -
versions— at least one version entry with:-
version— version string -
date— date in YYYY-MM-DD format -
models— at least one model withname -
scores— a list of score objects. Fields differ by benchmark:arc-agi-1 and arc-agi-2:
benchmark—"arc-agi-1"or"arc-agi-2"score— required; number 0–100set— required; eval set used (e.g."public","semi-private","private")cost— optional; USD cost to achieve score (positive number)scorecard_url— not allowed; scorecards are only available for arc-agi-3
arc-agi-3:
benchmark—"arc-agi-3"scorecard_url— required; link to your scorecard onthree.arcprize.org. arc-agi-3 scores are pulled automatically from the scorecard and should not be self-reported.set— required; eval set used (e.g."public","preview")cost— optional; USD cost to achieve score (positive number)score— not allowed; do not include a numeric score for arc-agi-3 entries
Using a list allows multiple entries for the same benchmark (e.g. different sets or cost tiers). Example:
scores: - benchmark: "arc-agi-2" score: 62.3 set: "public" cost: 1.50 - benchmark: "arc-agi-2" score: 45.0 set: "semi-private" cost: 1.50 - benchmark: "arc-agi-3" scorecard_url: "https://three.arcprize.org/scorecards/abc123" set: "preview" cost: 8.50
-
citation— how to reference your workpaper_url— link to a paper (must resolve if provided)twitter_url— link to a tweet or thread (must resolve if provided)
- Your
code_urlmust point to a public repository at the time of submission. - You may include additional files in your submission directory (READMEs, diagrams, etc.) but please keep it lightweight — no binaries, model weights, or large data files. Link to those instead.
- arc-agi-1/arc-agi-2 scores are self-reported. Misrepresenting results undermines the community and may result in your entry being removed.
- arc-agi-3 scores are pulled automatically from your scorecard — do not self-report a numeric score.
- Be respectful in discussions and PR comments.
submissions/
.example/
submission.yaml # Template — copy this
my-method-name/
submission.yaml # Your submission
README.md # Optional: extra detail about your method
another-method/
submission.yaml
Submissions are merged if they:
- Pass all automated checks
- Look like a genuine submission (not spam, not obviously fake)
- Have a working link to a public code repository
- Include a reasonable description of the method
We do not verify scores or run your code. ARC Prize may choose to verify submissions independently at a later time.
Open an issue or ask in the ARC Prize Discord.