Anthropic Ships Fable 5.1 With 75% Cheaper Cache Reads and Fewer False Refusals

Anthropic released Claude Fable 5.1 and Mythos 5.1 on September 1, making the pair generally available across the Claude API, AWS, Google Cloud, and Azure.

The headline changes are economic and behavioral rather than raw capability:

• Pricing holds at Fable 5's $10 per million input tokens and $50 per million output — but cache reads drop 75 percent, from $1 to $0.25 per million tokens, a significant saving for agentic workloads that repeatedly re-read long contexts.
• Fewer false positives: Anthropic retuned safeguards to reduce spurious refusals, one of the most persistent developer complaints on Fable 5.
• Multimodal input (text and images) is standard, and Anthropic self-reports a Terminal-Bench-Science score of 52.6, more than double Fable 5's 24.7.

The two models are identical under the hood; they differ in safeguards. Fable 5.1 is the generally available tier, while Mythos 5.1 runs with reduced restrictions and is offered only through trusted-access programs designed for cybersecurity and life-sciences work — Anthropic's answer to the capability-gating question OpenAI faced days later when GPT-6 Astra crossed its "Critical" cyber threshold.

Coming the same week as Astra's launch and Z.ai's aggressively priced GLM-5.3-Flash, the release underlines where frontier competition has moved: not just peak intelligence, but serving costs and the terms of access.