Anthropic launched Claude Fable 5.1 on Sept. 1 with higher coding and science scores and an API restriction designed to block a documented model-distillation technique.
Key Points:
- Fable 5.1 keeps $10/$50 pricing while cache reads drop 75% to $0.25 per million tokens.
- Anthropic reports gains in science, coding and business workflow benchmarks.
- New API accounts face a context-editing restriction aimed at blocking reasoning extraction.
Claude Fable Launch
Anthropic released Fable 5.1 alongside Claude Mythos 5.1, describing both as its most advanced systems for coding and knowledge work. Fable 5.1 is generally available. Mythos 5.1 remains limited to vetted users.
Input and output pricing stays at $10 and $50 per million tokens, while cache reads dropped 75% to $0.25 per million tokens. Anthropic estimates the lower cache price cuts typical workload costs about 25% and highly agentic workloads by as much as 45%.
Fable 5.1 scored 52.6% on Terminal-Bench-Science 0.1, up from Fable 5’s 24.7%, while its Terminal-Bench 4.0 coding score reached 55.8%. That coding result topped Opus 5 at 52.3% and OpenAI’s GPT-5.6 Sol at 37.3%, while AutomationBench rose to 31.4% from 17.1%. ClaudeDevs said its writing style is “more natural” and it can work longer before requesting input.
Also Read: Saylor Turns Buyer Again With 4,603 Bitcoin After A Summer Of Selling
Anthropic Copycat Defense
The launch also changes how API accounts created from Sept. 1 can handle Claude’s stored reasoning during multi-turn conversations, closing a route Anthropic says distillers used. The company said users can no longer “manually edit Claude’s prior context ... while preserving the transcript of Claude’s prior thinking.”
That restriction targets distillation, a common training method that can also be used to train cheaper rival models on outputs from more capable systems. Existing API accounts are not affected yet. Anthropic said the change will apply to all users with future model releases as it expands anti-distillation controls.
The move follows Anthropic’s February disclosure that DeepSeek, Moonshot AI and MiniMax generated more than 16 million Claude exchanges through roughly 24,000 fraudulent accounts.
Moonshot accounted for more than 3.4 million exchanges. Anthropic said the campaigns targeted capabilities including agentic reasoning, coding, tool use, data analysis and computer interaction.
Fable 5.1 extends the model line Anthropic introduced in June, while keeping core token prices unchanged and raising several benchmark scores over its predecessor. Its reliable knowledge cutoff is June 2026, and Anthropic says retirement will come no sooner than Sept. 1, 2027.
Read Next: Dyson Aims AI At Your Teeth With A $499 Camera Toothbrush And Jet
