Opus 5 improved its ARC-AGI 3 benchmark performance from under 5% (Opus 4.8) to over 30% in just two months. The model also emerged as more efficient than its predecessor, Opus 4.8, and offers comparable intelligence to Fable 5 at 26% lower Cost per Task, according to the ArtificialAnlys Benchmark.
Rapid model iteration cycles are driving significant performance and efficiency gains, intensifying the competitive pressure on other frontier models.