Voices
ARC-AGI-3 Launches March 25 — Chollet's First Interactive Reasoning Benchmark Uses Game-Like Environments
The ARC Prize Foundation is launching ARC-AGI-3 on March 25, 2026, the first major format change since the benchmark launched in 2019. Unlike ARC-AGI-1 and -2 (static visual reasoning), version 3 uses video-game-like environments where agents must explore without instructions, discover rules, plan across multiple steps, and adapt to novel hand-crafted levels that cannot be memorized. This directly tests whether frontier models have achieved the interactive, adaptive intelligence Chollet has consistently argued current LLMs lack.
Source
↳ Follow the thread