Anthropic Finally Unleashes Its Most Powerful AI Yet—With Guardrails

Anthropic has dropped a new AI model it’s calling Claude Fable 5, the company’s most capable system to date. But what makes this release unusual is that the model belongs to a family Anthropic once deemed too risky to share with the public.

Meet the Mythos class. Fable 5 is the first widely available model from this line, which the company previously held back over fears it could be misused for cybersecurity attacks. Now, Anthropic says it’s found a way to let the model loose—with limits. When Fable 5 encounters high-risk topics like advanced biology or hacking, it steps aside and hands the wheel to Claude Opus 4.8, a model the company describes as more cautious. In testing, 95% of sessions ran entirely on Fable without needing that fallback.

On performance, Anthropic claims Fable 5 outpaces competitors in software engineering, knowledge work, and visual tasks, especially as projects grow longer and more complex. The company also quietly released Claude Mythos 5, which it says is the same model but with some safety restrictions removed. For now, only organizations already in Anthropic’s private Project Glasswing program can upgrade to Mythos 5. Broader access is promised later through a more formal trusted-user system.

Pricing is steep: $10 per million input tokens and $50 per million output tokens—double the cost of Opus 4.8, though half the price of the earlier Mythos Preview. Anthropic didn’t explain why both models carry a “5” label or how they relate to the existing Mythos Preview. The company declined to comment on the record.