Anthropic has launched Claude Fable 5, the first publicly accessible version of its powerful Mythos model. While excelling in software engineering and knowledge work, this AI comes with strict safety limits, particularly in high-risk areas like cybersecurity.

Initially restricted to select partners due to security concerns, Anthropic is now rolling out access through subscriptions and Enterprise plans. Fable 5 will be included at no extra cost until June 22, after which usage credits are required. The company aims to reinstate it as a standard feature soon.

A new version of Mythos, dubbed Mythos 5, is being deployed to approved organizations, while Anthropic also stress-tested its classifiers to prevent jailbreaks. Concerned about potential misuse, the firm now requires a 30-day retention on all traffic, a policy that could set an industry precedent.

In third-party tests, Fable 5 outperformed other models in complex tasks and showed strong judgement. Pricing remains high at $10 per million input tokens and $50 per million output tokens, reflecting the advanced reasoning skills required for such tasks.