Anthropic releases Claude Fable 5.1 and Mythos 5.1: twice as good at science tasks and cheaper to run
Prepared by Sintelligent with the help of SI. Publication time restored from the build log. . Updated: .
The strongest public model has become noticeably smarter at scientific work and cheaper at the same time, so the long autonomous tasks SI does for hours are becoming more accessible to everyone.
Fable 5.1 scores 52.6% on Terminal-Bench-Science (Fable 5 had 24.7%), 55.8% on Terminal-Bench 4.0 and 60.9% on Humanity's Last Exam without tools. Cache reads are 75% cheaper (down to $0.25 per million tokens), which cuts the bill by about 25% on typical tasks and up to 45% on agentic ones; the base prices of $10/$50 per million tokens are unchanged. Mythos 5.1 is the same model with different safeguards, available only to vetted cybersecurity and life-sciences specialists in the US. The safeguards now trigger needlessly less often: 60% fewer false blocks in cybersecurity and 85% fewer on harmless biology queries. ARC Prize confirmed 97.5% on ARC-AGI-1 and 90% on ARC-AGI-2.