Anthropic launches Claude Fable 5.2 and Mythos 5.2, up to 30% faster at lower prices

Anthropic on Monday (November 9) launched Claude Fable 5.2 and Claude Mythos 5.2, saying the models improve coding, reasoning and vision tasks, generate output up to 30 percent faster than the 5.1 versions, and use fewer tokens on challenging tasks at medium effort. The models went live the same day on Claude paid plans, Claude Code, the Claude Platform and Amazon Web Services, Google Cloud and Microsoft Azure. (Image: file image of the Claude logo.)

Same underlying model, different access

Fable 5.2 and Mythos 5.2 share the same underlying model and differ in safeguards and access. The public Fable model carries cybersecurity and biology safeguards against misuse; Anthropic says it blocks about 55 percent fewer requests than Fable 5.1. Mythos has fewer limits and is especially strong at finding and reproducing software vulnerabilities; it is available only to organizations admitted through Anthropic’s Life Sciences and Cyber Verification Programs, and through Claude Security on Enterprise plans.

Faster and cheaper

Standard API prices are $8 per million input tokens and $40 per million output tokens, lower than the previous versions. An optional fast mode promises token generation up to 2.5 times faster, priced at $16 per million input tokens and $80 per million output tokens. Subscribers on eligible paid plans get extra Fable usage this week, with one to three resets depending on subscription length, and Fable accounts for up to 50 percent of weekly usage limits on Pro plans and standard Team seats. The speed and price figures are Anthropic’s; real-world results will vary by task and deployment.

Built for the hardest open-ended technical work

Anthropic recommends starting with Opus 5.5 for most everyday coding and knowledge work, switching to Fable or Mythos for more demanding requests, and stepping down to the more affordable Sonnet or Haiku 5.5 where appropriate. Company benchmarks say Fable and Mythos 5.2 outperform Opus 5.5 in computer use, knowledge work, cybersecurity, agentic scientific research, agentic coding and multidisciplinary reasoning; those results have not been independently verified.

Science claims need careful reading

The launch also carries two striking science claims from Anthropic: pre-release versions completed calculations addressing two open questions in theoretical black-hole physics, with preprints published; and about 1,000 model agents formalized the existing proof of Fermat’s Last Theorem in Lean, a language in which a computer can check each step, in 30 hours with minimal human intervention, against 11 days for an earlier research version. These are company claims and preprints, not yet broadly and independently verified; formalization turns an existing proof into machine-checkable form and is not the discovery of a new proof.

The trade-off over capability boundaries

The launch also raises an increasingly important question for frontier AI companies: how much of a model’s capability should be available to everyone. Anthropic manages the trade-off with its verified-access scheme — Mythos is open to defensive researchers but limits some legitimate uses, while Fable’s safeguards will block many sensitive cyber tasks. For buyers, the near-term story is a faster, cheaper frontier model with a clear access boundary; for scientists, the real test comes next: whether the formalized mathematics and physics results withstand close review, and whether comparable gains appear outside the company’s own demonstrations.