← Back to glossary
+Suggest a term
Concept·AI Models & Capabilities·Added today

Mythos-class

Also known as: Mythos class model, capability-gated model, safety-tiered model

A designation for AI models whose raw capabilities are judged too dangerous for general release, requiring safety-scoped variants to be made public instead. Coined around Anthropic's Claude Mythos, where Fable 5 was released as the general-access version and Mythos 5 kept behind a trusted-access gate.

When Anthropic announced Claude Mythos, it was notable not just for capability claims but for the decision framework around release. The company determined that the underlying model posed serious risks in the wrong hands, particularly around cybersecurity, and chose to release a safety-scoped derivative called Fable 5 for general use, while Mythos 5 itself was made available only to vetted cyberdefense and critical infrastructure providers. This introduced new vocabulary: a Mythos-class model is one that clears some capability threshold where the standard release calculus breaks down.

The concept matters for the builder community because it represents a formalized version of something that had been ad-hoc before. Labs have always made deployment decisions, but Mythos-class introduced the idea of a public taxonomy: there is a raw model, a safety-scoped public version, and a restricted-access version for trusted partners. Builders using Fable 5 are effectively using a curated subset of Mythos-level capability.

Practically, this is likely to become a recurring pattern. As frontier models grow more capable, the gap between what a model can do in research settings and what is safe to expose via a public API will widen. Understanding that a product you are building on may be several layers removed from the underlying research model, and that those layers are deliberate safety choices, is increasingly important context for anyone making technical bets on AI infrastructure.

This definition is AI-generated and refreshed weekly. It may contain inaccuracies. Use your own judgment, especially for production decisions.
Related terms
Claude MythosAlignmentModel safety evaluationRed teamingFrontier model