Claude Fable 5 and Mythos 5: Anthropic's next frontier release

Anthropic released Claude Fable 5, the public face of its new Mythos-class model. The unguarded sibling, Claude Mythos 5, stays restricted to cybersecurity partners and a small group of biology researchers. They share the same underlying weights and differ only in safety configuration. Pricing is set at $10 per million input tokens and $50 per million output tokens, which Anthropic says is less than half the price of the earlier Mythos preview.

The release notes lean on concrete results rather than benchmark slogans. Stripe used Fable 5 to migrate a 50-million-line codebase in a day, work the team had estimated at two months. On Hebbia's senior finance benchmark and Cognition's FrontierCode evaluation, Fable 5 took the top score. In long-context play, persistent file memory made the model three times more effective at Slay the Spire than Opus 4.8. Anthropic also claims Mythos 5 produced novel molecular biology hypotheses that domain scientists preferred to Opus-class output about 80 percent of the time in blind comparisons, and ran a week of autonomous genomics research that beat a recent Science paper while training a model a hundred times smaller.

The safety story is unusual. Cybersecurity, dual-use biology, and large-scale distillation attempts route to safeguard classifiers that fall back to Opus 4.8. Anthropic says these classifiers trigger in under five percent of sessions and openly admits some benign requests will get caught. Over a thousand hours of external red-teaming did not find a universal jailbreak.

Why it matters

If you ship with Claude, the pricing alone will move workloads: a faster model at lower per-token cost makes long agent loops viable that were borderline on Opus 4.8. If you work in research areas the safeguards touch, expect more silent fallbacks to the older model, and budget for that in your evaluations.