Anthropic recently released a new artificial intelligence model, Claude Fable 5, that is the first publicly available version of what the company calls its 'Mythos class' of models. An earlier, unrestricted preview of the Mythos model was shared with a limited group in April and was found to uncover thousands of software vulnerabilities.
Unlike the preview version, Anthropic says Fable 5 is being released with classifiers that detect certain types of requests and redirect them to a less powerful model, Claude Opus 4.8. The restricted categories include cybersecurity, advanced biology, chemistry, and attempts to extract the model’s training data.
When a request falls into one of the above areas, the user receives a response from Opus 4.8 instead. Dianne Penn, Anthropic’s head of product management for research and labs, told Reuters: “Let’s say I’m a college student asking the model, ‘help me find cyber vulnerabilities on X package or code.’ The model would refuse, and Fable 5 will fall back to Opus 4.8 for a response.”
According to Anthropic, the classifiers trigger in fewer than 5% of sessions on average. The company said it conducted extensive testing, including an external bug bounty programme that ran for over 1,000 hours without producing a universal jailbreak. Independent red-teaming organisations also failed to find a universal jailbreak on long-form agentic tasks, though Anthropic noted that the UK’s AI Safety Institute had made progress towards one in a short testing window.
Alongside Fable 5, the company launched Claude Mythos 5, the same underlying model but with cybersecurity restrictions removed, for a small group of pre-approved partners including US government entities under the Glasswing programme. Anthropic plans to expand access to Mythos 5 through a broader trusted access programme. A separate programme for biology and chemistry researchers is also being developed.
Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens, less than half the cost of the earlier preview. Penn told Reuters that the model uses fewer tokens per task, which lowers the overall cost.
Fable 5 is available immediately via the Claude API and consumption-based enterprise plans. Subscription users on Pro, Max, Team, and seat-based Enterprise plans can access it at no extra charge until June 22, after which usage credits will be required. Anthropic said it aims to restore the model to standard subscription plans when capacity allows.
The launch takes place as Anthropic, valued at $965 billion, moves towards an initial public offering alongside rival OpenAI. The decision to release the model with visible restrictions reflects the company’s strategy of pairing high-capability systems with what it describes as safety measures.
