Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on 1 September, and the headline number is not a benchmark. It is $0.25 — the new price of a cached input token, down from $1.00 on the previous Fable release.
Two names, one model
Fable 5.1 and Mythos 5.1 are the same underlying system shipped under different safeguard regimes. Fable 5.1 is generally available. Mythos 5.1 is restricted to vetted organisations working in cybersecurity and the life sciences, reached through a trusted-access programme rather than the open API.
That split is unusual. Most labs ship one model and gate capabilities inside it. Anthropic is instead gating who can reach the less-restricted configuration at all, which makes the access programme itself part of the product.
The pricing change is the story
Standard pricing is unchanged: $10 per million input tokens and $50 per million output tokens. Batch processing halves both, to $5 and $25.
What moved is the cache. A cache read now costs $0.25 per million tokens — 2.5% of the standard input price, where most Claude models charge 10%. Anthropic puts the practical effect at roughly 25% lower cost on typical workloads and up to about 45% on context-heavy agentic ones.
For anything that re-reads a large context on every turn — an agent working through a repository, a long-running research task, a document pipeline — that is the difference between a cost line that scales with turns and one that mostly does not.
Benchmarks, and what they do not settle
On Terminal-Bench-Science 0.1, an agentic scientific research benchmark, Anthropic reports Fable 5.1 at 52.6%, against 29.0% for Opus 5, 24.7% for Fable 5 and 22.4% for GPT-5.6 Sol. On Terminal-Bench 4.0, Fable 5.1 reaches 55.8% and Mythos 5.1 reaches 60.9%.
The company also lists 73.4% on CursorBench 3.2.0, 31.4% on AutomationBench, 41.7% on the strict split of OSWorld 2.0 and 77.9% on the partial split, 1,853 on GDPval-AA v2, and Humanity’s Last Exam at 60.9% without tools and 65.0% with them.
Every one of those figures comes from Anthropic’s own evaluation of its own model. They are worth reading as a claim about direction, not as an independently reproduced result. The gap between Mythos 5.1 and Fable 5.1 on Terminal-Bench — five points on identical weights — is the clearest signal in the table, because it measures what the safeguard regime costs rather than what the model knows.
Where it runs
The model answers to claude-fable-5-1 in the API, and ships across Claude Code, Claude Enterprise and the Claude Platform. Cloud availability covers Amazon Bedrock, Google Agent Platform and Microsoft Foundry.
Anthropic did not publish a context window figure for either model.
What to watch
Two things. First, whether independent evaluators reproduce the Terminal-Bench-Science gap — a 23-point lead over the previous generation is large enough that outside confirmation matters. Second, whether a 2.5% cache multiplier holds. It is priced well below the rest of the Claude line, and repricing a cache is easier than repricing a model.
