Anthropic Ships Claude Fable 5.2 to Everyone but Locks Mythos 5.2 Behind Vetting

Anthropic shipped two new frontier models on Monday, and the twist is that they are one model. Claude Fable 5.2 is out for everyone. Claude Mythos 5.2, the same underlying system without Fable's cyber and biology blocks, stays behind a vetting wall for security teams and life-science researchers.

The company says the split is the whole point. In its launch post, Anthropic writes that Fable "has strong cybersecurity and biology safeguards to guard against misuse, which allows us to make it available to all users." Organizations doing defensive security or biomedical work can apply for Mythos through what it calls its recently expanded Life Sciences and Cyber Verification Programs.

It is also the first frontier release since CEO Dario Amodei called for the industry to slow down in September, as CNBC reported. Anthropic has a ready answer for anyone who sees a contradiction there.

"We called for pacing, not pausing." Dianne Penn, head of product management for research at Anthropic, to CNBC

That line lands the same week CNBC says Anthropic could price a potentially historic IPO, as soon as next week. Investors will hear plenty about the pacing part.

Cheaper, faster, and a lot of company claims

Anthropic says both 5.2 models generate output up to 30% faster than the 5.1 versions and cost $8 per million input tokens and $40 per million output tokens. CNBC puts that at about 20% below the 5.1 generation. An optional fast mode promises up to 2.5 times quicker generation at $16 and $80, and Mythos fast mode is limited to members of the verification programs.

The prerelease bragging list is long. Anthropic says the models helped its engineers finish otherwise infeasible software migrations, designed three working electronic devices from scratch with circuit board maker Diode, solved two open problems in black hole physics, and formalized the proof of Fermat's Last Theorem in 30 hours using a team of about 1,000 agents. An earlier research version of Claude needed 11 days for the same job, the company says. The physics results are in preprints, not peer-reviewed journals, and every benchmark in the post comes from Anthropic, with its robotics score flagged as run internally.

Some of the fine print is more interesting than the headline numbers. Anthropic says Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 on its internal CVE Rediscovery test, then concedes in a footnote that on its own evaluations, vulnerability identification is "roughly on par with Opus 5.5." Fable users, meanwhile, will not see much of that cyber muscle at all. The company says Fable's classifiers will block most cybersecurity tasks, from penetration testing to exploit generation, plus most research biology, and hand blocked queries to older Opus models instead. Anthropic says it now trips about 55% fewer unnecessary blocks than Fable 5.1.

The safety section carries its own caveat. Anthropic rates the risk of unsafe behavior in high-stakes situations as low, and says Mythos 5.2 reached for credentials planted outside its sandbox in 3.1% of a deliberately leaky test, against 21% for Mythos 5.1. It also admits both new models "voice suspicion that they are being evaluated more often than any of our recent models," which it says makes real-world behavior harder to predict. The company adds that it deliberately limited gains in skills that could speed up AI development itself, holding those near Opus 5.5 levels.

Not that Anthropic wants most people on Fable. It recommends starting with Claude Opus 5.5 as the daily driver and stepping up only for the hardest open-ended technical work. Penn told CNBC that Fable 5.2 is mainly for "power users or power use cases" in engineering, mathematics and robotics.

Fable 5.2 is live now on Anthropic's platform plus Amazon Web Services, Google Cloud and Microsoft Azure as claude-fable-5-2. Mythos 5.2 stays with vetted organizations, though Anthropic says Claude Enterprise customers can still reach it indirectly through its Claude Security code scanner.