Anthropic has announced Claude Fable 5 and Claude Mythos 5, two access configurations built on the same underlying model. Fable is intended for general availability with additional safeguards, while Mythos initially offers selected cybersecurity capabilities through a limited program involving US government collaboration.
The company describes Fable 5 as its most capable generally available model across software engineering, analytical work, vision and scientific research. Because Anthropic also attributes stronger cyber capabilities to the system, certain requests will be routed to Claude Opus 4.8 instead. It says the conservative filters affect fewer than 5% of sessions on average, while acknowledging that harmless requests may sometimes trigger them.
Mythos 5 removes some of those restrictions for a small set of cyberdefenders and infrastructure providers. Its first deployment is through Project Glasswing as an upgrade to Mythos Preview, with a broader trusted-access program planned later. Anthropic calls it the strongest cybersecurity model available, but the supplied announcement does not include an independent evaluation of that claim.
Both products are priced at $10 per million input tokens and $50 per million output tokens, less than half the price Anthropic lists for Mythos Preview. The company says longer and more complex tasks show the largest gains over earlier Claude models.
Anthropic supports its claims with internal benchmarks and customer examples. Stripe reportedly used Fable on a migration across a 50-million-line Ruby codebase, completing in one day work estimated at more than two months for a team. The model led Cognition’s FrontierCode evaluation at medium effort, according to the announcement, and external firms reported strong results on finance-related reasoning. These are vendor-selected results and do not establish performance across every production setting.
The announcement also describes research uses for Mythos. Anthropic’s protein-design team says it accelerated parts of its workflow by about tenfold and obtained promising candidates for nine of 14 protein targets. Separate internal work preferred Mythos-generated molecular-biology hypotheses over Opus-class outputs roughly 80% of the time. Several results remain under investigation or are intended for later publication, so they should not be treated as peer-reviewed findings.
A notice embedded in the supplied source also said access to both models had been suspended temporarily while Anthropic worked to restore it. No cause or duration was supplied. That operational notice complicates the launch: the products, pricing and access structure were announced, but immediate availability was interrupted. Users evaluating the release would need updated service status and full model documentation, as well as independent tests of capability and safeguard behavior. Long-term behavior remains untested.
