One model, two ways to get access

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9. The company says they use the same underlying model. What changes is how it is offered: Fable applies additional checks to sensitive requests in cybersecurity and biology, while Mythos is available through more restricted routes.

Fable is available through Claude and Anthropic’s developer platform, with Anthropic also listing AWS, Google Cloud and Microsoft Azure. Cybersecurity users and life-science researchers can apply through separate verification programs. Claude Enterprise customers can also use the Claude Security scanning product, which Anthropic says can use Mythos 5.2. That is not the same as direct access to the model.

The access route matters

Anthropic’s cyber program describes tiers with different conditions and permitted work. Individuals can apply for the defense tier; higher tiers are for organizations. Approval is not a general license to test somebody else’s systems. Teams still need to know who can use the model and which work they are authorized to do.

Anthropic lists standard API rates of $8 per million input tokens and $40 per million output tokens for both versions. Tokens are the units used to meter the text and other information a model processes or produces. An optional fast mode costs twice as much per token where available.

What early testing can tell us

Anthropic reports roughly 55% fewer safeguard blocks for Fable 5.2 than for Fable 5.1 in early tests across its own products. That figure counts blocks in the company’s tests. It does not show how often a harmful request will get through, or how frequently legitimate work will be blocked in a given organization.

The practical comparison is therefore not only which model performs better on a test. It is which version a team can use, under what rules, and how the safeguards behave on the tasks it is allowed to run.

Sources

  1. Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2November 9 launch, shared underlying model, availability, pricing and company-reported safeguard tests.
  2. Claude Help Center: Cyber Verification ProgramPrimary access documentation describing tier conditions and individual versus organization eligibility.