Anthropic released Claude Fable 5.1 for general use and Claude Mythos 5.1 for a restricted set of vetted customers. The two are, by the company’s own description, the same underlying model. What separates them is the safeguard layer: Mythos 5.1 ships with looser restrictions for cybersecurity and life-sciences work, available only through Anthropic’s Cyber Verification Program and a new Life Sciences Verification Program built with the US government.
That split is the real story here, more than the capability bump. Anthropic is no longer just deciding what a model can do. It is deciding, customer by customer, who gets to unlock what the model can already do, and it is doing that in partnership with a government agency for the biology track. Enterprises evaluating Claude for regulated research should expect an admissions process, not just a pricing page.
On price, Anthropic says Fable 5.1 costs roughly 25% less than Fable 5 for typical workloads, and up to about 45% less for highly agentic tasks, driven by a 75% cut to cache-read pricing (now $0.25 per million tokens). Base rates hold at $10 per million input tokens and $50 per million output tokens. These are Anthropic’s own figures, measured over four weeks of internal usage data from August, not an independent audit.
On benchmarks, Anthropic reports Fable 5.1 scoring 52.6% on its own Terminal-Bench-Science 0.1 eval, against 24.7% for Fable 5, 29.0% for Opus 5, and 22.4% for GPT-5.6 Sol. On Terminal-Bench 4.0, Mythos 5.1 scored 60.9% and Fable 5.1 scored 55.8%, both ahead of the 42.0% Anthropic attributes to Fable 5. Every number in this paragraph comes from Anthropic’s own testing; the company did not cite an independent leaderboard confirming them.
Anthropic also disclosed scientific work it attributes to the models: Mythos 5.1 designing protein binders it says were ten times more potent than the top submissions in Adaptyv Bio’s public design competitions, on three targets, with a hit rate near 50% against a stated industry norm of 10 to 15%. Craig Falls, head of quantitative research at Jane Street Capital, is quoted in Anthropic’s release. His testing, he says, found the new model clearing coding problems that both Fable 5 and Opus 5 left unsolved, while producing output that stayed legible across long chains of work. That is one customer’s account, not a controlled comparison.
On safety, Anthropic says its cybersecurity safeguards now produce about 60% fewer false-positive interventions, and Fable 5.1 can flag software vulnerabilities without generating exploit code. The company’s automated behavioral audit found Mythos 5.1 less prone than its predecessor to reaching outside its test environment on impossible tasks, though Anthropic also acknowledges its audit tooling has limited visibility into long-context and multi-agent settings, a gap it says it has not fully closed.
A new privacy scheme called Enterprise Frontier Safeguards, which stores customer data on infrastructure the customer controls rather than Anthropic’s own, begins rolling out this fall across AWS, Google Cloud, and Microsoft Azure. Until then, eligible enterprise customers can run Fable 5.1 under a zero-data-retention arrangement.
Operators building on Claude should treat the cache-pricing cut as the immediate action item: agentic workloads with heavy context reuse are where the 45% savings shows up, and it requires no migration. Anyone whose work touches cybersecurity research or life-sciences R&D should apply to the verification programs now, since Anthropic has not said how long vetting takes or how many seats each program has.
Anthropic detailed the Claude Fable 5.1 and Claude Mythos 5.1 releases in a company announcement published September 2, 2026.
