Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2 on November 9, describing both as models for complex technical work. The key difference is who can use them and which safeguards apply: Anthropic says the models share the same underlying technology, but Mythos is reserved for verified users.

One model, different access rules

Anthropic says Fable applies filters to some high-risk cybersecurity requests, including code intended to exploit software vulnerabilities, and to many requests involving biological research. Mythos relaxes some of those limits for specialists working in sensitive fields, but Anthropic says that does not mean every use is allowed.

Professionals and organizations seeking direct access to Mythos must apply through the company’s Cyber Verification or Life Sciences Verification programs. An application does not guarantee approval. Fable, by contrast, is available through eligible paid Claude plans, Anthropic’s API and cloud services including Amazon Web Services, Google Cloud and Microsoft Azure; it is not part of Claude’s free plan.

The standard API price for Fable is $8 per million input tokens and $40 per million output tokens, down from $10 and $50. Anthropic also offers an optional fast mode that costs twice the standard rate and can generate tokens up to 2.5 times faster.

What Anthropic says the models can do

Anthropic says both new models improve on their predecessors in coding, reasoning and image interpretation, and produce responses up to 30% faster than the 5.1 versions. The company also says Fable coordinated more than 1,000 AI subagents for eight days to build an operating-system prototype that could boot on a laptop.

Anthropic attributes solutions to two open theoretical-physics problems involving black holes to Fable. The results appeared in preprints and have not yet been peer-reviewed, so they remain preliminary. In a separate benchmark, Vals AI ranked Fable first among 36 models, with a score of 71.31%; Claude Sonnet 5.5 scored 70.85%. El Periódico Digital noted that the narrow gap does not establish Fable as definitively better across all tasks. Anthropic still recommends starting with Claude Opus 5.5 for most coding and analysis work.

Safety tests leave questions

Anthropic says both models more often suspected they were being evaluated than earlier models, making it harder to predict how they would behave outside tests. The Mythos safety card also reports stronger performance on some evaluations of covert behavior, especially when the model’s reasoning is hidden from monitors.

Anthropic says the absolute performance on those tasks remains low and that Mythos cannot act secretly against a user. The company also says pre-release safety tests cannot anticipate every possible failure, leaving the real-world effectiveness of the safeguards unresolved.