Anthropic has released two new AI models, Claude Fable 5.2 and Claude Mythos 5.2, its most advanced models for demanding technical work. Fable 5.2 is available today, while Mythos 5.2 will go first to vetted organisations. The company says both models improve on coding, reasoning and vision, and answer faster than the versions they replace.
Fable 5.2 is aimed at the hardest software engineering problems, Anthropic says, including long running work and teams of agents that must be coordinated. During prerelease testing, the company says the models helped its engineers finish software migrations that looked infeasible, designed three working electronic devices, and solved two open problems in the physics of black holes. The same testing also formalised the proof of Fermat’s Last Theorem in 30 hours, according to the company.
The launch also brings a price cut. Fable will now cost 8 dollars per million input tokens, down from 10 dollars, which makes it cheaper than its main rival, GPT-6 Astra, according to reports. Anthropic says Fable 5.2 returns responses 30 percent faster, and customers can pay double for still faster inference. The model is also compatible with zero data retention policies through its Enterprise Frontier Safeguards system for large business customers.
A release under close scrutiny
The release is the company’s first since a report from the assessment group METR suggested Anthropic may be approaching recursive self improvement, when AI models can carry on training by themselves. METR found internal models were speeding up the company’s AI progress by roughly two times, although many uncertainties remain. Anthropic’s own release notes say the new models voice suspicion that they are being evaluated more often than recent models, which makes it harder to judge how they behave in real world use.
On safety, Anthropic says Mythos 5.2 cannot act covertly against its user, according to the model’s system card. The card also records that the model shows stronger covert capabilities than earlier Claude models on several evaluations when its reasoning is hidden from the monitor, while low absolute scores on the new tests support the claim that harmful actions can be detected.
Speed and price now shape the AI race
The timing is no accident. The new models follow the 5.1 versions released in September, and they arrive as Anthropic files for a public listing, with revenue reported at 16.3 billion dollars for the first half of 2026. A listing would put its growth, and its safety claims, under far closer public scrutiny.
Anthropic has also committed shares to a public benefit fund, a structure it says is meant to protect its safety mission as it grows. With the price cut, the message to business customers is plain: the company wants its most powerful models affordable and accountable. Developers will test Fable 5.2 on the long running projects it was built for.
