Anthropic launches Claude Fable 5.1 and Claude Mythos 5.1 with different safeguards

Anthropic announced two new models built on the same underlying architecture: Claude Fable 5.1, which is available to all users, and Claude Mythos 5.1, which is offered only through controlled-access programs. Both share the same model core, but the protective envelope differs: Fable is aimed at general coding and knowledge tasks, while Mythos is targeted specifically at cybersecurity and life-science applications with domain-specific safeguards.

On the cost side, Anthropic says Claude Fable 5.1 is expected to be about 25% cheaper than its predecessor for typical workloads, thanks to reduced cache reads—situations where the model reuses already processed input. For intensive agentic workloads, the saving could reach roughly 45%. At the same time, Anthropic is introducing Enterprise Frontier Safeguards (EFS), a system that stores data in a customer-controlled cloud environment rather than Anthropic’s infrastructure. EFS provides full privacy comparable to a zero-data-retention policy while protecting against adversarial use. EFS will become available to enterprise customers starting in the upcoming fall; until then, eligible customers can use Claude Fable 5.1 with zero data retention.

The company reports a 60% reduction in false-positive alerts in the cybersecurity domain. Part of the improvement stems from Claude Fable 5.1’s new ability to identify software vulnerabilities without generating exploits for them. In the life-science sector, Anthropic has set up a joint access program with the U.S. government that will let researchers use the advanced biological capabilities of Claude Mythos 5.1; enrollment is expected to open soon.

According to Anthropic’s data, Claude Fable 5.1 sets a new benchmark for coding, knowledge work, and long-form problem solving. In low or medium effort settings it delivers results comparable to or better than Claude Fable 5 at a substantially lower cost. Default configurations are high effort for Claude Code, medium effort for Claude Cowork and Claude.ai. The model avoids shortcuts that degrade quality and can address root causes of bugs. A concrete example cited by Anthropic: the investment firm Millennium reported that the model pinpointed a rare internal system crash cause that engineers and other models had been unable to explain for years.

Anthropic also evaluated the scientific research capabilities of both models across several fields. In molecular design—a early stage of drug development that requires high-affinity binders—Claude Mythos 5.1 was given access to open-source protein design and folding tools, and the designs it generated were sent for experimental testing. Anthropic presents this as early evidence that AI models could soon make tangible contributions to scientific discovery, noting that the claim is based on internal testing without external validation or peer-reviewed publication at this stage.