Anthropic released two models on Monday: Claude Fable 5.2, available to all users, and Claude Mythos 5.2, restricted to vetted cybersecurity and life-sciences organisations. Both follow the 5.1 releases in September and are pitched at what the company calls "daunting" technical work.

The price is the part developers will notice first. Fable 5.2 lists at $8 per million input tokens and $40 per million output tokens — roughly 20% below the previous generation — and Anthropic says it generates tokens up to 30% faster, with a Fast mode promising up to 2.5x. It ships across the major clouds, including AWS.

The capability claims are unusual even for a launch day. Anthropic says the models, during prerelease testing, helped engineers finish software migrations that had been judged infeasible, designed three working electronic devices from scratch, solved two open problems in the physics of black holes, made "substantial progress" on robotics evaluations, and formalised the proof of Fermat's Last Theorem in 30 hours using 1,000 agents.

Coding, reasoning and vision are the advertised gains. Fable 5.2 is positioned for long-running projects and for orchestrating teams of agents rather than answering single prompts. Mythos 5.2 keeps the gated distribution of earlier Mythos releases: vetted organisations only, on the argument that its cyber and biology capabilities need controlled access.

The interesting signal is the shape of the release rather than any single benchmark. A price cut, a speed bump and a gated sibling on the same day is the pattern of a lab competing on cost per finished task — and it landed hours before Meta put a 10-cent-per-million-token model on the market for the repetitive work agents do in between.

Whether the physics and mathematics claims survive independent scrutiny is now the open question. Anthropic has published benchmark tables; it has not published the problem statements or the agent transcripts behind them.