Anthropic launches Claude Fable and Mythos 5.2 models

Anthropic says Claude Fable 5.2 and Claude Mythos 5.2 formalized a proof of Fermat’s Last Theorem in 30 hours during prerelease testing, a result the company presented as evidence that its newest models can sustain long, difficult technical work.

What does Anthropic claim the models can do?

Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2, describing the pair as models built for “daunting” technical challenges. The company says they improve on the 5.1 generation in coding, reasoning and vision, and produce output faster than Fable 5.1 and Mythos 5.1.

Fable carries the clearest software-engineering emphasis. Anthropic says it can handle challenging engineering problems, orchestrate teams of agents and carry work through long-running projects. Anthropic also says Fable has solved complex scientific problems, including two open theoretical-physics problems.

Anthropic described the prerelease results this way:

“During prerelease testing, the two models have helped our engineers complete otherwise-infeasible software migrations, designed three working electronic devices from scratch, solved two open problems in the physics of black holes, made substantial progress on robotics evaluations, and formalized the proof of Fermat’s Last Theorem in 30 hours.”

Those examples shift the proposed role beyond a one-shot code answer. A migration becomes an end-to-end assignment; device work becomes a hardware design task; scientific work becomes an attempt to resolve open questions. In each case, Anthropic describes the models directing several operations toward a finished or advanced result.

The measurable outputs are specific: three electronic devices designed from scratch, two black-hole physics problems solved, and a proof formalized within 30 hours. The 30-hour figure works like a stopwatch for one claimed task, not a benchmark across models. The company also reports substantial progress on robotics evaluations. That gives prospective users concrete tasks to test, rather than a general promise that the models are more capable.

How strong is the evidence for faster output?

The speed claim remains less specific than those task outcomes. Anthropic’s announcement provides no benchmark names, scores, sample sizes, latency measurements or test conditions for its assertions about coding, reasoning and vision. It does not quantify the improvement in output speed over 5.1 either.

For developers, that omission matters. A model can generate tokens more quickly while still taking longer to complete a multi-step task, or it can finish a task sooner by taking more actions per minute. The announcement does not say which measure it used, so the practical distinction between faster drafting and faster delivery remains unresolved.

Nor does it explain the coordination method for agent teams, the progress checks or the verification of scientific outputs. Those omissions do not erase the claims, but they limit what readers can conclude from examples selected by the company.

Who can access the 5.2 models?

Fable 5.2 is available on launch day. Mythos 5.2 begins with a narrower audience: only vetted organizations will have access initially. Anthropic does not state what vetting requires, how organizations qualify or when broader access will follow.

The release follows Anthropic’s September launch of Claude Fable 5.1 and Claude Mythos 5.1, making improvement over those predecessors the central comparison for the new pair.

What should teams test?

For software teams, the decisive question is whether an orchestrated agent team can complete a bounded migration with less manual intervention than 5.1, not merely write a plausible patch. For device and robotics work, the relevant standard is whether the design functions when built and evaluated. The announcement supplies no pass rates, testing protocol or independent validation for either.

For research teams, a 30-hour formalization gives a duration but not a replication path. Readers need the inputs, proof framework, validation process and comparisons with 5.1 before they can judge whether the models can repeat the result.

None of that makes the announcement empty. Anthropic has named tasks and an elapsed time, giving the release more operational specificity than a generic capability claim. The next test is whether those outputs hold up outside the company’s prerelease examples and under the access conditions announced for Mythos.

As Fable reaches users and Mythos enters vetted organizations, reproducible results on those same tasks will determine whether Anthropic’s faster, more capable model claim changes real technical work.