Anthropic says Fable 5.2 agents formalised Fermat’s Last Theorem in 30 hours

Fable 5.2 and Mythos 5.2 are the same underlying model. Fable carries safeguards against cybersecurity and biology misuse and is available to all users; Mythos, with lighter restrictions, goes to organisations admitted to Anthropic’s cyber and life-sciences verification programmes. The company said the models generate output up to 30% faster than the 5.1 versions and cost $8 per million input tokens and $40 per million output tokens.

On FrontierCode 2.0, which asks whether a model’s code changes would be merged by the maintainers of open-source projects, Anthropic reported 49.1% for Fable 5.2, against 45.5% for Opus 5.5 and 42.8% for OpenAI’s GPT-6 Astra. On Humanity’s Last Exam with tools it reported 74.2%. Artificial Analysis, which runs its own tests, placed Fable 5.2 first of 74 models on its intelligence index with a score of 65, and also among the slowest and most expensive per task.

The result drawing most attention is mathematical. Anthropic said a team of about 1,000 Fable 5.2 agents fully formalised the proof of Fermat’s Last Theorem in Lean in 30 hours, starting from minimal prompts and a list of intermediate theorem statements, with no mathematical references. Its first formalisation, by an unreleased research model with a purpose-built harness and human guidance, took 11 days; the new proof is seven times shorter, and Opus 5.5 in the same setup completed 26% in 18 hours. The company also said the model settled two open problems about black holes in holographic duality, described in two preprints. None of these claims has yet been independently reviewed.