Anthropic Releases Public Version of Powerful Claude Mythos 5 Model With Heavy Guardrails on High-Risk Domains

Anthropic released Claude Fable 5 on Tuesday, making a Mythos-class model available to any user for the first time. The system uses the same underlying model as the restricted Claude Mythos 5 but adds classifiers that block or reroute queries involving cybersecurity, biology, and chemistry.

The company said the mechanism errs on the side of caution, so some benign requests may still be handled by a less capable model.

Claude Mythos 5 itself is being offered only to a narrow set of industry partners and select biology researchers. S. government is among the recipients and is also distributing the model through the Project Glasswing consortium, which first received Mythos capabilities last week.

Both new models carry identical pricing of $10 per million input tokens and $50 per million output tokens. That rate is twice the price of Anthropic’s earlier public models yet lower than the Mythos Preview tier released in April. Anthropic said more than 1,000 hours of red-teaming found no universal jailbreaks for the public Fable 5 version.

The company is still supplying unrestricted Mythos 5 instances to a handful of trusted partners while it develops a broader trusted-access program. Diane Penn, Anthropic’s head of Product Management, said the firm began addressing the model’s software-vulnerability discovery abilities before the April preview.

“We’re trying to make improvements in a way that’s beneficial, even if we don’t have the perfect solution for every use case to start,” she said.

Claude Fable 5 shows measurable gains on software-engineering and visual-understanding benchmarks compared with the April preview. Anthropic stated it continues to work on classifier precision and plans further releases once additional safeguards are ready.