Measuring Intelligence with Interactive Benchmarks: ARC Prize Foundation @ MIT
The ARC Prize Foundation is a non-profit for the public advancement of open artificial general intelligence (AGI). We drive progress through global competitions, research community events & resources, and open source benchmarks used by frontier labs like OpenAI and xAI.
Coming in '26, ARC-AGI-3 is the world’s first interactive reasoning benchmark: ~100 novel game environments where AI agents must explore, plan, and adapt to win - without instructions.
Designed to measure skill acquisition efficiency against human baselines, ARC-AGI-3 is the most authoritative test yet for progress toward AGI.
Join us on an exploration of how we can measure intelligence and the connections between how humans and AI play games and reason. In partnership with the Department of Brain and Cognitive Sciences at MIT.
Schedule
2:00pm - Doors, Play Games
2:15pm - ARC-AGI-3 Intro - Greg Kamradt
2:30pm - ARC-AGI-3, Background and rationale - Francois Chollet + Mike Knoop
2:45pm - Lightning Talks
4:00pm - Panel w/ Francois, Samuel, Laura, Jacob, and Josh
4:30pm - Wrap-up + Networking
ARC Prize Benchmarks
ARC-AGI-1 (Published)
Our inaugural eval resisted AI systems for 5 years even after multiple global, large-prize competitions, which is notable considering most evals saturate to human performance much faster.
ARC-AGI-2 (Published)
To increase signal toward general intelligence, this upgraded eval removes “low-signal” tasks. ARC-AGI-2 emphasizes tasks which require higher levels of abstraction + recombination.
ARC-AGI-3 (In Progress)
New format featuring game-like, interactive format. Learn more about ARC-AGI-3 or play the preview games.