GPT-6 ASTRA BEATS HUMANS ON ARC-AGI-3, PULLING AGI TIMELINE FORWARD

Astra's performance splits evaluators. Epoch AI ranks it first with 169 points, while Artificial Analysis places it behind Claude Fable 5.1 and level with its predecessor GPT-5. The divergence highlights ongoing inconsistencies in AI benchmarking standards. The breakthrough comes on ARC-AGI-3, a test designed to measure abstract reasoning. Astra's ability to match human performance efficiency represents a significant milestone for the benchmark's developers. Chollet stopped short of declaring AGI achieved, but noted progress is accelerating. He estimates development is proceeding twice as fast as his previous projections, leading him to move forward his AGI timeline accordingly. The discrepancy between benchmarks underscores a persistent challenge in AI evaluation: different testing frameworks often yield conflicting conclusions about model capabilities. As frontier models advance, standardized, reliable metrics remain elusive.