ARC-AGI

This app contains the visual puzzles of the ARC-AGI benchmark.

Each puzzle follows a simple, hidden rule. Your task is to discover the rule and solve the puzzle. These are not tests of knowledge or memory. They are tests of fluid intelligence: the ability to reason and solve new problems.

The ARC-AGI benchmark was created to provide a fair comparison between human and artificial intelligence. By using puzzles that do not rely on prior knowledge, it measures a machine's ability to learn and adapt. To date, no AI has been able to solve all of these puzzles.

The app lets you engage with the very same problems researchers use, giving you a hands-on sense of what remains difficult for machines.