Anthropic has released Claude Fable 5.2 and Claude Mythos 5.2, a pair of frontier models aimed at the hardest technical work. The two share one underlying model and differ mainly in safeguards: Fable carries strong cybersecurity and biology guardrails so it can ship to all users, while organisations doing beneficial cyber or biomedical research can request Mythos access through Anthropic's expanded Life Sciences and Cyber Verification Programs.
The numbers move in almost every category Anthropic tested. Fable 5.2 scores 69.9% on FrontierSWE v2 (up from 56.3% for Fable 5.1), 74.2% on Humanity's Last Exam with tools, 89.3% on OSWorld 2.1 computer use, and 73.4% on Chartography visual chart recognition. It also generates output up to 30% faster than 5.1 and costs less: USD 8 per million input tokens and USD 40 per million output tokens, with an optional fast mode that roughly 2.5x's token generation for USD 16 and USD 80.
Anthropic says prerelease testers used the models to finish software migrations that were otherwise infeasible, design three working electronic devices from scratch, solve two open problems in black hole physics, and formalise the proof of Fermat's Last Theorem in 30 hours. The launch lands in a crowded field alongside GPT-6 Astra and Gemini 4 Carbon, and independent tracker Artificial Analysis scored Fable 5.2 at 65, the highest it has recorded. Anthropic's own guidance is measured: it still recommends Opus 5.5 as the daily driver for most coding and knowledge work, reserving Fable and Mythos for ambitious, open-ended projects.