Anthropic launched Claude Fable 5 today, making its most capable generation of AI available to paying subscribers and enterprise customers for the first time. Before now, the underlying Mythos technology was locked behind a program called Project Glasswing, accessible only to a small group of vetted organizations including Amazon, Apple, Cisco, Google, Microsoft, and JPMorganChase.
Fable 5 runs on the same core model as Mythos 5. The difference is a set of automatic filters layered on top. When a question touches cybersecurity, biology, chemistry, or what Anthropic calls "distillation" scenarios (attempts to copy the model's knowledge into a competing system), the filter intercepts the request and hands it to Claude Opus 4.8 instead. Anthropic says this happens in fewer than 5% of sessions.
Why so cautious? The answer comes from what Project Glasswing partners found when they actually used Mythos. In roughly one month, about 50 organizations collectively found more than 10,000 high-severity or critical security flaws in widely-used software. One partner bank used it to catch a fraudulent $1.5 million wire transfer. Cloudflare found 2,000 bugs across its critical systems, with a false-positive rate better than human testers. The model is genuinely powerful at finding and explaining how to exploit software weaknesses.
The practical concern is that this power cuts both ways. A model that helps a bank's security team find vulnerabilities can, in the wrong hands, help an attacker exploit them instead. Anthropic has said publicly that it expects competing AI companies to release Mythos-class models within six to twelve months, possibly without these kinds of restrictions. The current window is their attempt to make sure defenders are better prepared before that happens.
For most business users, the restrictions will be invisible. Software engineering, document work, data analysis, research, and customer-facing tasks run entirely on Fable 5 without any fallback. The 5% threshold means a company would have to be doing very specific technical or scientific work before hitting the guardrails regularly.
Pricing dropped significantly with the public launch. Access via the API is now $10 per million input tokens and $50 per million output tokens, down from $25 and $125 respectively during the restricted Mythos Preview period. Through June 22, Fable 5 is included at no extra cost on Pro, Max, Team, and enterprise plans. After that, Anthropic plans to move it to usage credits before working to restore it as a standard subscription feature.
All of this is happening as Anthropic filed a confidential IPO prospectus with the SEC on June 1, with a valuation reported at roughly $965 billion. Revenue has grown from about $9 billion at the end of 2025 to a reported $47 billion annualized run rate by late May 2026. The Fable 5 launch is not a coincidence in that context. A broadly available, flagship product matters for the story Anthropic needs to tell public market investors.
For business operators, the question is simple: does your work involve asking an AI about technical vulnerabilities, biological processes, or chemical information? If yes, you will occasionally get a less capable response than you might expect from the model you are paying for, and the system will tell you when that happens. For most other professional tasks, Fable 5 is the most capable publicly accessible AI model Anthropic has ever released.