Anthropic releases Claude Fable 5.1, more than doubling predecessor on key benchmark

Anthropic released Claude Fable 5.1 on September 1, 2026, more than doubling its predecessor's score on the Terminal-Bench-Science 0.1 benchmark with 52.6%. While base pricing remains at $10/$50 per million tokens, a 75% cut to cache read costs reduces agentic workload expenses by up to 45%. To combat model copying, Anthropic blocked new API accounts from editing conversation history to preserve Claude's stored thinking, targeting distillation campaigns linked to Chinese labs like Moonshot AI.

Claude Fable 5.1 release

• ▪Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026, marking the first update to its Mythos-class model line since June 9, 2026
• ▪Claude Fable 5.1 is generally available to anyone with a Claude account, while Claude Mythos 5.1 is restricted to vetted cybersecurity and life-sciences professionals
• ▪Claude Fable 5.1 and Claude Mythos 5.1 are built on the same underlying model but utilize different safety filters

Benchmark performance improvements

• ▪Claude Fable 5.1 scored 60.9% without external tools and 65.0% with them on Humanity's Last Exam, outperforming Claude Opus 5
• ▪Claude Fable 5.1 scored 55.8% on Terminal-Bench 4.0, outperforming Claude Fable 5 at 42.0%, Claude Opus 5 at 52.3%, and OpenAI's GPT-5.6 Sol at 37.3%
• ▪Claude Fable 5.1 scored 52.6% on Terminal-Bench-Science 0.1, more than doubling the 24.7% score of its predecessor, Claude Fable 5
• ▪Claude Fable 5.1 scored 31.4% on AutomationBench, nearly doubling the business workflow scores of its predecessor

Cache cost reductions

• ▪Anthropic reduced API cache read costs by 75% to $0.25 per million tokens, while keeping base pricing unchanged at $10 input and $50 output per million tokens
• ▪The 75% reduction in API cache read costs cuts typical workload costs by roughly 25% and highly agentic workloads by up to 45%

Anti-distillation protections

• ▪Anthropic blocked new API accounts opened on or after August 31, 2026, from manually editing Claude's prior context in multi-turn conversations while preserving stored thinking
• ▪White House science adviser Michael Kratsios accused Moonshot AI of distilling Anthropic's Fable model to develop its Kimi K3 system
• ▪Anthropic traced over 16 million Claude exchanges to distillation campaigns run by roughly 24,000 fake accounts, naming Chinese labs DeepSeek, Moonshot AI, and MiniMax

Subscription tier access structure

• ▪Claude Fable 5.1 is not included in standard Pro plans or standard Team seats, which must run the model using pay-as-you-go usage credits
• ▪Claude Fable 5.1 is included in Max plans and premium Team or Enterprise seats for up to 50% of weekly usage limits

2 sources