Anthropic introduces Claude Fable 5.2 and Mythos 5.2 models

Both versions are built on a shared base model. Fable 5.2 is open to all users, but it includes additional guardrails against high-risk tasks in cybersecurity and biology. Mythos 5.2 is available only to organizations that have been verified by Anthropic, with applications accepted through the Cyber Verification and Life Sciences Verification programs.

In the announcement, the company provided examples of their performance. One thousand Fable 5.2 agents translated the proof of Fermat's Last Theorem into the Lean language, spending 30 hours on the task; Anthropic's September experiment with another model took 11 days. In another test, Fable 5.2 spent eight days coordinating more than a thousand sub-agents that built a prototype operating system from scratch that boots on a laptop.

In the FrontierSWE v2 benchmark, Fable 5.2 scored 69.9% compared to 62.3% for Opus 5.5. In the internal CVE Rediscovery benchmark, Mythos 5.2 identified 89.6% of vulnerabilities, while Opus 5.5 with guardrails disabled found 81.3%. In a sandbox escape test, Mythos 5.2 used external data in 3.1% of scenarios, compared to 39% for Mythos 5 and 21% for Mythos 5.1.

The base pricing for both models is $8 per million input tokens and $40 per million output tokens. Fast Mode speeds up generation by up to 2.5 times and costs $16 and $80, respectively. Fable 5.2 is already available in Claude services, via API, and on cloud platforms including AWS, Google Cloud, and Microsoft Azure.