Benchmark Almanac

ARC Prize Foundation

We track 3 benchmarks and 1 arena that ARC Prize Foundation made or helped make.

3 in use 4 with public code 4 with a leaderboard here

Arena leaderboards

Top five in each of its 1 arena with three or more rated models, the most rated first.

Benchmarks

Name Kind Released Best score Status
ARC-AGI-2 Models infer a rule from example grid pairs and produce the output grid for new inputs. Scored by tasks solved within two attempts. Abstract puzzles and logic Mar 2025 95.8% In use
ARC-AGI Few-shot grid puzzles: infer a hidden transformation from about three examples and produce the output grid. Semi-private sets verified by ARC Prize. Abstract puzzles and logic Nov 2019 98.5% In use
ARC-AGI-3 Interactive game environments where agents must explore, infer goals and rules, and act efficiently without instructions. Puzzles and chess Mar 2026 100% In use