Announcing ARC-AGI-2 and ARC Prize 2025
- Source
- arcprize.org
- Author
- Greg Kamradt
- Date
Context
ARC Prize designed ARC-AGI-2 to expose reasoning weaknesses while checking that its puzzles remained solvable by people. For its 2025 launch, each task had to be solved by at least two human participants within two attempts. That is a check on the tasks: it does not mean every participant solved every puzzle.
The new tasks targeted difficulties such as applying several interacting rules at once, reading symbols for their meaning rather than their visual pattern, and applying a rule differently depending on . The launch also paired scores with cost per task. Together, those choices make the ask how efficiently a system adapts to unfamiliar problems, rather than simply whether enough computation can eventually produce an answer.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
Checking sign-in…
Loading comments…