ARC-AGI-4 Benchmark for Autonomous Open-Ended Innovation Scheduled for Q1
September 12, 2026
The ARC Prize team is developing ARC-AGI-4 to measure autonomous open-ended invention in AI models. The benchmark aims to track progress in meta-skills required for scientific innovation rather than standard pattern matching.
HOW THIS AFFECTS YOU
●
researcherYou can use this new metric to evaluate model capability in reasoning and novel problem-solving beyond static datasets.