ARC Prize Verified (formerly ARC-AGI-Pub) is a secondary leaderboard measuring the ARC-AGI public evaluation set. No competitions or prizes are associated with this leaderboard.
This leaderboard imposes no limitations on internet access and few limitations on compute. That means you can utilize state-of-the-art models.
Selected high score claims are verified by ARC Prize using a semi-private evaluation set. New high scores are accepted when there is agreement between the two scores (+/- 10%).
Only new high score claims (e.g., submissions that exceed the current high score by at least 1%) are eligible be verified and published. This is to limit the semi-private evaluation set from leaking into LLM training data.
See more details on the official website.
Please allow 2 weeks to verify submission.