Anthropic launched Claude Fable 5.1 for general customers on 1 September and Claude Mythos 5.1 for vetted organizations in cybersecurity and life sciences. The company says they are the same underlying model; differences on some tasks mainly reflect tighter safeguards on the public version. Fable may help discover vulnerabilities but not develop exploits, while Mythos remains in a restricted access program.

Fable 5.1's base price remains $10 per million input tokens and $50 per million output tokens. Anthropic has, however, reduced the price of reading previously cached context. Based on four weeks of its own usage, it estimates about 25% savings for typical workloads and up to roughly 45% for context-heavy agentic work. That estimate depends on the share of cache reads and is not a blanket reduction for every token.

In Anthropic's evaluations, Fable 5.1 scored 52.6% on Terminal‑Bench Science 0.1 and 55.8% on Terminal‑Bench 4.0; Mythos reached 60.9% on the latter. The tests used production safeguards, and some have modified settings, so cross-company comparisons need care. The company also reports 60% fewer false-positive safety blocks in cybersecurity tasks.

Anthropic also published experiments in which Mythos designed protein binders and optimized computing kernels. A near-50% binder hit rate across twelve targets and up to tenfold better affinity on three targets were lab-validated within the company-described project, but this is not yet an independent peer-reviewed demonstration. The system card also concedes that Mythos is slightly worse than Claude Opus 5 on overall misaligned behaviour, although better than the previous Mythos 5.