ChainThink reports that on June 10, according to an official announcement, Anthropic launched its new flagship large model, Claude Fable 5, the first Mythos-level model made publicly available, with full API and enterprise edition access now live.
The model is priced at $10 per million input tokens and $50 per million output tokens, more than halving the price of the test version. Simultaneously launched is the upgraded Claude Mythos 5 with the same architecture, designed for security-focused organizations with fewer restrictions.
In terms of performance, Claude Fable 5 scored 29.3% on the FrontierCode Diamond difficulty benchmark, significantly outperforming GPT-5.5 at 5.7% and Claude Opus at 13.4%.
Stripe's test showed that it can migrate a 50-million-line Ruby codebase in a single day—a task that previously required a team two months to complete.
Additionally, the model can pass the game Pokémon FireRed via screenshots, reconstruct web source code, and predict adenovirus capsid assembly through reasoning alone in biology, outperforming specialized protein models. It has independently completed single-cell genomics research and designed a classification model superior to those published in Science, with only one percent of the parameters.
To prevent misuse of advanced capabilities, Anthropic has deployed an independent safety classifier that automatically reverts to Claude Opus 4.8 when high-risk queries are detected, affecting fewer than 5% of sessions on average.
Enterprise traffic data will be retained for 30 days, and all manual reviews will be logged. Glasswing partners can remove the cybersecurity restrictions on Claude Mythos 5, and biomedical restrictions will be gradually lifted for specific institutions through a trusted access mechanism.
Subscribers can trial this model for free until June 22. After June 23, usage will require consumption points until computing power is sufficiently available to reinstate it under the standard subscription.