Pass IndexThe State of AISign in

ARC-AGI

Visual puzzles where each task follows a rule the solver has never seen, designed so memorisation cannot help.

how it is measured · the vocabulary

ARC prizeabstraction and reasoning

The benchmark most resistant to scale for the longest time, and therefore the one watched as a proxy for genuine reasoning rather than recall. Its prize competition publishes cost per task alongside score, which is unusual and useful.

Nearby

BenchmarkBenchmark contaminationEloEval setGPQAHumanEvalLLM as judgeLMArena