Anthropic reportedly splits Claude Fable and Mythos access by risk level

Anthropic has reportedly introduced Claude Fable 5.2 for broader use and Claude Mythos 5.2 for vetted users working in higher-risk fields such as cybersecurity and biological research. The distinction, according to Spanish technology publication Xataka, is not a different underlying model but different safeguards and access conditions. The details have not been independently confirmed in the material available for this draft, so users should check Anthropic’s own product and pricing pages before making purchasing or deployment decisions.

The report says Fable is available through Anthropic’s platforms and cloud providers, while Mythos requires approval through verification programs. If accurate, the split matters to developers and security teams weighing access to advanced technical capabilities against misuse risks. It also means “generally available” should not be read as free or unrestricted.

Key facts

Different access rules, not necessarily different intelligence

Xataka describes Fable and Mythos as versions of the same underlying model with different safety restrictions. It says Fable blocks some high-risk cybersecurity assistance, including code for exploiting vulnerabilities, and restricts many activities related to biological research. Mythos reportedly removes some of those limitations for approved users.

That distinction is important: the report does not establish that Mythos is generally more capable or that it permits every type of request. Its stated difference is access to certain capabilities under a verification process. For developers, the practical question is therefore not only which model performs better, but which tasks each version allows—and under what terms.

Availability and reported pricing

According to Xataka, Fable can be used through Anthropic’s services and providers including Amazon Web Services, Google Cloud and Microsoft Azure. Access through Claude depends on the subscription plan and is not included in the Free plan, the report says. It lists API pricing of $8 per million input tokens and $40 per million output tokens, and mentions temporary conditions for some paid subscribers without detailing them in the supplied material.

These terms can change, and the article does not provide a direct link to an official pricing page. Before budgeting or integrating the API, customers should verify current prices, plan eligibility, regional availability and any provider-specific terms directly with Anthropic or the cloud vendor.

Mythos access reportedly requires approval

The report says that Mythos is not available simply by paying for a subscription or API usage. Interested professionals and organizations must apply through Anthropic’s Cyber Verification or Life Sciences Verification programs, and an application does not guarantee access. Xataka also describes a route for Claude Enterprise customers to use Claude Security, which it says incorporates Mythos for code analysis. That is not the same as unrestricted access to the model.

For security teams, the reported arrangement makes eligibility and scope central procurement questions. They should confirm what data the service processes, which features are enabled and what contractual and privacy conditions apply; those details are not established by the source material provided here.

Performance claims need context

Xataka attributes several demonstrations to Anthropic, including an experiment in which Fable coordinated more than 1,000 subagents over eight days to build a prototype operating system. The company is also said to have reported that the model solved two open theoretical-physics problems concerning black holes. The report notes that the results appeared in preprints, which are not equivalent to peer-reviewed validation.

The article also cites a Vals AI index ranking Fable first among 36 models at 71.31%, ahead of two named competitors by narrow margins. Such a ranking is evidence about performance on that index’s specific tasks, not proof of broad superiority. The supplied material does not include the benchmark’s methodology or a direct results link, so the score should be treated as a reported claim rather than a definitive comparison.

What readers should verify next

The central news is the reported policy choice: wider access to one version, with more restrictive access to another for sensitive work. Whether the safeguards reliably prevent harmful use while allowing legitimate research is not answered by launch claims alone. Independent testing and clear documentation of access controls will matter as much as headline benchmark scores.

Before adopting either model, check Anthropic’s official model documentation, current pricing, verification requirements, safety documentation and terms. The source material for this draft does not include those primary documents, so model names, commercial details and capability claims remain attributed to Xataka rather than independently verified.

Source: Xataka IA, “Anthropic acaba de lanzar sus dos modelos de IA más avanzados. Uno tendrá acceso general y el otro estará restringido”: https://www.xataka.com/robotica-e-ia/anthropic-acaba-lanzar-sus-dos-modelos-ia-avanzados-uno-tendra-acceso-general-otro-estara-restringido