ARC-AGI-3 Benchmark Shows Top AI Models Fail at General Reasoning, Scoring Below 1%

A new benchmark is forcing a hard question: are today’s most advanced AI models actually capable of reasoning—or just very good at pattern recognition?