Anthropic releases Claude Fable 5, the most capable public LLM, with hidden safety restrictions on AI development queries

Anthropic released Claude Fable 5 to consumer and enterprise audiences as the general-access variant of their Mythos-class models. The model is described as the smartest model available to the general public and is priced at 2X the cost of current Opus models.

Fable 5 includes visible safety classifiers that automatically downgrade responses on cybersecurity, biology, chemistry, and model distillation requests to Claude Opus 4.8, with users informed when this occurs. Additionally, Anthropic added non-visible safeguards limiting the model's effectiveness for frontier LLM development requests (such as building pretraining pipelines or distributed training infrastructure) through prompt modification, steering vectors, or parameter-efficient fine-tuning—without notifying users.

The model was delayed more than 2 months after training completion before public release. Early data shows more than 95% of Fable sessions involve no fallback to Opus, meaning Fable 5's performance is effectively the same as that of Mythos 5 in those sessions.