Anthropic on Tuesday launched the latest versions of its flagship Fable and Mythos models, slashing cache read pricing by 75% to $0.25 per million tokens. The company says the updated models deliver stronger performance at a lower cost than their predecessors, with Fable 5.1 now generally available while Mythos 5.1 remains restricted to its trusted access program for cybersecurity and life sciences work. Core pricing holds steady at $10 for input and $50 for output per million tokens, but the cache price cut represents the headline improvement for users who rely on repeated context.
In benchmarks shared by Anthropic, Fable 5.1 outperforms the older Fable 5 model, Anthropic's Opus 5, and OpenAI's GPT-5.6 Sol across nearly all categories. The most dramatic gain appears in the Terminal-Bench-Science 0.1 benchmark, which tests the model's ability to use agents for scientific research — Fable 5.1 scored 52.6% versus Fable 5's 24.7%. Most other improvements fall within a 2% to 4% range compared to either Fable 5 or Opus 5, according to the company. Anthropic says users can step down the reasoning mode by one or two levels and still match Fable 5 results at higher settings, using fewer tokens overall. Because of this efficiency, Fable 5.1 defaults to high reasoning mode in Claude Code but uses medium settings on Claude Cowork and Claude.ai.
The company says it has made the model's safeguards "more precise, ensuring that they're less likely to flag benign content (like queries about medical issues or cyber defenders using the model to make their systems safer), but still ensuring they provide robust protection against genuine threats." Cyber safeguards now intervene about 60% less per session than with Fable 5, and the model can now identify vulnerabilities in code — though penetration testing, exploit generation, and binary scanning remain off-limits. Biology safeguards intervene 85% less often on routine questions about basic biology and medical topics. Anthropic is also opening a detection API in private preview for eligible organizations to check for watermarks in text generated by models released after August 2, 2026, as required under EU law.
The update addresses enterprise privacy concerns that limited Fable 5 adoption in regulated industries. Anthropic is rolling out Enterprise Frontier Safeguards, a system that stores data in cloud infrastructure controlled entirely by the customer — such as Amazon S3, Azure Blob Storage, or Google Cloud Storage — encrypted with the user's own keys. The company runs automated detection against that data with no human review on Anthropic's side and routes any alerts back to the customer. The system will launch in phases starting this fall, with no additional cost beyond the customer's cloud storage bill. Until then, Anthropic won't retain any data from Fable 5 or Fable 5.1 usage. The company says it designed the solution after businesses in regulated sectors proved unable to use the model under the previous 30-day retention period. Fable 5.1 also ships with strengthened mechanisms to block distillation — Anthropic has closed off a common technique by preventing new accounts from manually editing Claude's prior context in multi-turn conversations while preserving the transcript of Claude's prior thinking. The restriction will expand to existing accounts with future model releases, though it may disrupt agent harnesses that rewrite or compact conversation history. For enterprises weighing reasoning models, the shift from mandatory retention to customer-controlled storage could prove as consequential as the performance gains themselves. The watermark detection API, meanwhile, signals a pragmatic path through compliance requirements that might otherwise have fragmented deployment across jurisdictions.