Technology

ARC-AGI-3

An interactive reasoning benchmark designed to measure an AI agent's fluid intelligence and skill acquisition efficiency in novel, rule-free environments.

Launched in March 2026 by the ARC Prize Foundation, ARC-AGI-3 is the first fully interactive benchmark in the ARC-AGI series, specifically engineered to expose the gap between human fluid intelligence and AI pattern matching. The framework drops AI agents into hundreds of handcrafted, turn-based game environments featuring zero instructions, explicit rules, or stated goals. To succeed, agents must actively explore, form hypotheses, and adapt on the fly using only core knowledge priors. While human test-takers solve these dynamic environments with a 100% success rate (averaging a median time of 7.4 minutes), leading AI models struggle to clear even 1%, highlighting a critical bottleneck in modern artificial general intelligence development.