Anthropic has released Claude Fable 5, its first generally available Mythos-class model, which means Mythos is now public except for the parts that made Mythos a geopolitical object in the first place.

The company announced Tuesday that Fable 5 is the most capable model it has ever made widely available. Anthropic says it exceeds prior Claude models across software engineering, knowledge work, vision, scientific research, and long-running autonomous tasks. It is also built on the same underlying model as Claude Mythos 5, the restricted model Anthropic is giving to a small group of cyberdefenders and infrastructure providers through Project Glasswing.

The distinction is the safeguards. Fable 5 is the public model. Mythos 5 is the same model with safeguards lifted in some areas. When Fable 5 receives queries in high-risk domains like cybersecurity, biology, chemistry, and model distillation, Anthropic says the system can route the request to Claude Opus 4.8 instead. The company says those safeguards trigger in fewer than 5 percent of sessions on average, which is another way of saying that 95 percent of the time users get the Mythos-class model and 5 percent of the time they get the model Anthropic thinks is safer.

That is a product split and a governance framework at the same time.

For months, Mythos has been the model that Anthropic said was too capable to release broadly because of its cyber abilities. It was first distributed through Project Glasswing, where vetted partners used Claude Mythos Preview to scan codebases for vulnerabilities. Anthropic later said those partners had found more than 10,000 high- or critical-severity security flaws. Then came the access politics: US agencies wanted it, the UK evaluated it, the EU pushed to get ENISA into the program, and Anthropic expanded Glasswing to roughly 150 additional organizations across more than 15 countries.

Anthropic has turned that political compromise into a product architecture. A public version for everyone, a restricted version for trusted users, and a classifier layer deciding when the public version quietly stops being the public version.

The pricing tells the same story. Anthropic is offering Fable 5 and Mythos 5 at $10 per million input tokens and $50 per million output tokens. That is more expensive than Opus 4.8, according to The Verge, but less than half the price of Mythos Preview, according to Anthropic. The model is being commercialized, not merely studied.

The capability claims are large. Anthropic says Fable 5 can work autonomously longer than previous Claude models and that its advantage grows as tasks become longer and more complex. Stripe reported that Fable 5 completed a codebase-wide migration in a 50-million-line Ruby codebase in a day, work Anthropic says would otherwise have taken a team more than two months. Anthropic also says Mythos 5 accelerated parts of internal protein-design work by around 10 times and generated molecular-biology hypotheses its scientists preferred to Opus-class outputs in blinded comparisons roughly 80 percent of the time.

Vendor claims, all of them, until independent evidence accumulates. But the structure of the release already matters on its own. Anthropic has moved from “this is too dangerous to release” to “this is safe to release if a classifier redirects the dangerous parts.” That is a genuinely different theory of frontier AI governance. It does not require withholding the whole model. It requires slicing capability by use case, user trust, and safety routing.

Cybersecurity is dual-use by design. A vulnerability-finding model helps defenders because it can find vulnerabilities. The same property makes it useful to attackers. Anthropic’s answer is to sell the model broadly, reserve the less-restricted version for trusted users, and let safeguards police the boundary. Whether that boundary holds is a different question.

So Mythos is public now. Also Mythos is still restricted. The difference is Fable, which is a model name, a pricing tier, and a theory of how to release dangerous capability without quite releasing all of it.

–
By the Control Plane Editorial Team