Did Anthropic Just Have Twins? Claude Fable 5.1 and Mythos 5.1

What Anthropic shipped

Anthropic released two models with one name and two surnames: Claude Fable 5.1 and Claude Mythos 5.1. Anthropic describes Fable as a Mythos-level model and as its most capable generally available model. Fable is on the Pro, Max, Team and Enterprise plans, on the Claude Platform, and through cloud marketplaces. Mythos is not for sale. It goes only to vetted organizations through Anthropic’s trusted access programs, starting with cyber defenders in the Cyber Verification Program.

Same model, different guardrails

The twins share the same underlying model. What differs is the set of safeguards wrapped around it. Fable carries what Anthropic calls robust safeguards for cybersecurity and biology. When a request lands in one of those areas, Anthropic routes it to a less capable model instead of letting Fable answer at full strength. Mythos keeps the same capability but drops those specific restrictions, which is why access is limited to teams whose work is exactly that: finding and fixing vulnerabilities, or doing life-science research under supervision.

So the honest answer to “which one is smarter” is neither. The difference is who is allowed to point the full model at the two areas Anthropic considers dual use.

Why split a model in two

The same skill that lets a model find a flaw in a codebase lets it help someone break in. Anthropic’s answer is to keep one model and ship two policies. Everyone gets the reasoning, coding, vision and long-running agent work. The narrow slice that could do real damage in the wrong hands is fenced off behind verification, rather than removed from the model entirely or handed out to everyone.

That is a different bet from training a smaller “safe” model for the public. Fable is not a cut-down sibling. Outside the two fenced areas, it is the same thing Mythos users get.

What this means if you build with Claude

If you are building agents, coding tools or research workflows, Fable 5.1 is the model to target. Anthropic positions it for ambitious, long-running, asynchronous work, and it is priced at $10 per million input tokens and $50 per million output tokens, with cache reads at $0.25 per million. If your product touches offensive security or biology, expect Fable to route those requests to a weaker model, and plan for it: either apply to the trusted access program or design around the boundary.

One thing Anthropic has not published as of this writing is a context window figure for either model, so check the platform docs before you size a prompt budget.

Sources: for this explainer we checked Anthropic’s Claude Fable 5.1 page.