OutYet reporting
Anthropic’s Fable 5.1 Makes Safeguards Part of the Model Comparison
Fable 5.1 and restricted Mythos 5.1 share the same underlying model, while different cyber safeguards affect which tasks they can complete. Anthropic’s own evaluation describes both the gains and the limits of its comparisons.
Anthropic introduced Claude Fable 5.1 for general availability and Claude Mythos 5.1 for restricted access on September 1. The company says they use the same underlying model, with Mythos 5.1 available through vetted access programs, including for cybersecurity and life sciences work. That distinction matters to anyone comparing capability claims: the two names do not describe separate base models, but users may still encounter different behavior because safeguards intervene differently.
The update follows the original Fable 5 and Mythos 5 launch, and Anthropic frames the 5.1 versions as improvements for coding and knowledge work. Its announcement includes comparisons on Terminal-Bench 4.0 and Humanity’s Last Exam, including results at different costs and tool settings. These are useful details for developers choosing a model, but the scores come from Anthropic’s own evaluation rather than an independent, like-for-like test. TechCrunch independently reported the general-availability versus restricted-access split and described the pair as versions of the same underlying model.
One particularly useful caveat appears in Anthropic’s Terminal-Bench discussion: the company attributes the performance gap between Fable 5.1 and Mythos 5.1 to tasks where earlier cyber safeguards intervened. Anthropic says it expects the gap to shrink as those safeguards improve. That means a benchmark result can reflect both the model’s underlying ability and the policy layer applied to a given version. Comparing a Fable result with a Mythos result without accounting for those interventions can therefore give a misleading impression of the difference between the models.
For technical users, the practical question is not just which model scores higher, but which access path and safeguards fit the task. Fable 5.1 is the broadly available option described in the announcement; Mythos 5.1 remains limited to vetted programs. Anthropic also cautions that its alignment assessment has limits, including gaps in evaluation coverage, so its results do not settle how either model will behave across every workflow. The published evidence supports a more specific conclusion: safeguard design is part of the product comparison, and benchmark tables should be read with that design in view.