Anthropic releases Fable 5.2 with faster, cheaper reasoning

Anthropic released Claude Fable 5.2 and Mythos 5.2 on November 9, targeting demanding technical projects. The two configurations share an underlying model; Fable includes cybersecurity and biology safeguards for broad access, while Mythos requires approval through Anthropic’s verification programs. Anthropic still recommends starting with Opus 5.5 for most everyday work.

A still from the electronics design demonstration accompanying the release. Source: Anthropic.

Standard pricing falls from $10/$50 to $8/$40 per million input/output tokens, a 20% reduction. Anthropic reports up to 30% faster output than version 5.1; an optional fast mode promises up to 2.5 times the generation speed at $16/$80. The developer documentation lists a one-million-token context window, 128,000-token maximum output and always-on adaptive thinking.

The company reports stronger coding, reasoning and vision results, including 49.1% on FrontierCode 2.0 at its Xhigh effort setting. Its launch thread describes 1,000 agents formalizing Fermat’s Last Theorem in 30 hours. That means translating an existing proof into computer-checkable form, rather than discovering the theorem’s proof anew; it is also a coordinated multi-agent result rather than a single chat response.

Anthropic’s reported proof-formalization progress: Fable 5.2 reaches completion at 30 hours. Source: Claude’s launch thread.

Early X reactions point to different parts of the release: Dan McAteer highlights the price reduction and efficiency, while Haider draws attention to Anthropic’s stated limits on accelerating AI research. These are initial reactions, not independent replications of the launch results.

Sources: Announcement, Original X post, Developer documentation, Formalization thread, Dan McAteer, Haider