Fable 5.1 & Mythos 5.1: Same Model, Different Modes

Anthropic announced the release of Claude Fable 5.1 and Claude Mythos 5.1, which are identical models with different levels of safety filtering. Fable 5.1 is publicly accessible today at a reduced price of $10/million tokens of input and $50/million tokens of output, with cached memory reads being 75% cheaper (which usually corresponds to ~25% cost savings, or up to 45% savings in agentic scenarios). Safety filters have also been refined so that they are triggered 85% less frequently on benign biological and health-related topics, and interventions have been reduced by approximately 60% in the context of cybersecurity.

Mythos 5.1 uses the same weights but with fewer restrictions. It’s not sold publicly, however, and is only accessible via approved programs for infrastructure protection and life science research. Currently, it’s limited to a few US organizations. This shows that safety filters aren’t just based on the model’s capabilities, but rather who can be trusted with removing them.