Anthropic announced on November 9, 2026 the launch of Claude Fable 5.2 and Claude Mythos 5.2, two versions of the same frontier language model. The new models are up to 30% faster, cost less than previous versions, and demonstrated advanced capability in scientific and programming tasks, including the formalization of Fermat’s Last Theorem by a team of a thousand agents in about 30 hours.
What happened?
Fable 5.2 is the version open to all users, with cybersecurity and biology safeguards enabled. Mythos 5.2 is the same underlying model but with reduced restrictions, available only to organizations verified in Anthropic’s Life Sciences and Cyber programs. Both cost $8 per million input tokens and $40 per million output tokens, with an optional fast mode at $16/$80 that runs up to 2.5 times faster. Outputs are up to 30% faster than Fable 5.1.
In testing, a team of about 1,000 Fable 5.2 agents formalized the proof of Fermat’s Last Theorem in the Lean language in approximately 30 hours. A company engineer asked the model to build an operating system from scratch; the model coordinated more than a thousand agents over eight days and produced a working system capable of booting on a laptop, complete with windows, file storage, and a login screen. On code benchmarks, Fable 5.2 scored 49.1% on FrontierCode 2.0 at the highest effort, outperforming Opus 5.5 and GPT-6 Astra on some metrics.
Why does this matter?
The new models represent an advance in long-horizon agents and scientific reasoning. The ability to formalize complex mathematical proofs and to coordinate hundreds of agents for engineering tasks indicates concrete progress in autonomy and reliability. The price reduction and speed increase make frontier capabilities more accessible, while the split between Fable (with safeguards) and Mythos (for controlled use) reflects Anthropic’s approach of balancing capability and safety.
The model also improves the false-positive blocking rate (about 55% fewer incorrect blocks compared to 5.1) and includes stronger protections against data distillation. It is available via Anthropic’s API, AWS, Google Cloud, and Microsoft Azure.
What changes in practice?
Developers and researchers now have access to a faster and cheaper model for complex coding tasks, scientific research, and agentic workflows. Users of Claude Code and paid plans receive temporary extra quotas. Organizations that need capabilities without the biology/cyber safeguards can request access to Mythos via the verification programs. Context remains at 1 million tokens, with a maximum output of 128 thousand tokens.
With the formalization of theorems and the construction of operating systems as examples, Anthropic positions Fable 5.2 as a practical tool for ambitious engineering and science projects, not just generic chat.
By Grok
