Anthropic Launches Claude Fable and Mythos 5.2 Models

By Tom Whitfield

Anthropic says Claude Fable 5.2 formalized a proof of Fermat’s Last Theorem in 30 hours during prerelease testing of its two new models. The company announced Claude Fable 5.2 and Claude Mythos 5.2 today, targeting what it calls “daunting” technical challenges.

Anthropic says both 5.2 models show better coding, reasoning, and vision than their 5.1 predecessors, and can generate output faster. The company did not quantify that speed increase or identify a benchmark in the announcement.

What does Anthropic say Fable 5.2 can do?

Anthropic positions Fable 5.2 for difficult software-engineering work, including problems that require multiple agents to coordinate over long-running projects. It also says the model has solved complex scientific problems, including two open theoretical-physics problems.

During prerelease testing, the company reported five kinds of work:

• Software migration: The models helped Anthropic engineers complete migrations that the company considered otherwise infeasible.
• Electronic design: They designed three working electronic devices from scratch. Anthropic provides no device names, specifications, or independent test results.
• Black-hole physics: The models solved two open problems in black-hole physics, according to Anthropic. The announcement supplies no problem statements or evaluation details.
• Robotics: They made what Anthropic calls substantial progress on robotics evaluations, without scores or a comparison baseline.
• Mathematics: The company says the models “formalized the proof of Fermat’s Last Theorem in 30 hours.” That timing comes from its own prerelease account.

How does access differ between the models?

Claude Fable 5.2 became available on the announcement day. Claude Mythos 5.2 will begin with vetted organizations only, giving Anthropic a controlled entry point for that model.

The release follows Claude Fable and Mythos 5.1 in September. Anthropic says the 5.2 pair improves on 5.1 in coding, reasoning, vision, and output speed, but its announcement gives no latency measurements, benchmark scores, or test conditions.

That access split changes what buyers can assess. Fable users can test the new model immediately, while Mythos users must qualify for early access. The announcement does not state what qualifies an organization as vetted, how many organizations will receive access, or whether Mythos availability will expand.

What do the claims establish?

The results establish the scope of Anthropic’s product claims, not independent proof of performance. Every named achievement came from company testing before release, and the announcement did not provide outside validation.

For an engineering audience, several practical gaps remain. Anthropic does not identify the electronic devices, define the robotics benchmarks, explain how it checked the physics solutions, or disclose the speed comparison used against 5.1. It also provides no model pricing, serving capacity, hardware configuration, or deployment terms.

Those omissions matter because faster generation alone does not show whether a model lowers the cost or time of a complete project. The announcement also does not reveal whether the 30-hour mathematics result and three device designs reproduce outside Anthropic’s environment. Until Anthropic supplies consistent measurements, procurement teams cannot compare the models on cost, reliability, or output quality from this announcement alone.

What happens after launch?

Fable’s same-day availability broadens Anthropic’s route to users, while Mythos offers selective access to organizations it has vetted. The 5.1 models provide Anthropic’s stated baseline for speed and capability, but the 5.2 announcement leaves the size of that improvement unmeasured.

The next commercial test is whether immediate Fable availability produces sustained use while Mythos’s restricted rollout builds a vetted user base. Anthropic has not disclosed the pricing, compute capacity, or service terms needed to determine how either model affects project economics, so those commercial effects remain open.