Anthropic shipped Claude Fable 5 on Tuesday, the first general-availability model in its Mythos capability tier, alongside a restricted Claude Mythos 5 reserved for vetted cyber defenders inside Project Glasswing, the lab’s partnership with the U.S. government. The split release lets Anthropic claim frontier capability and gated dangerous capability in the same press cycle.

Fable 5 ships with classifiers that block high-risk cybersecurity, biology, chemistry, and distillation prompts, routing them to Claude Opus 4.8 as a fallback. Anthropic says the safeguards trigger in fewer than 5 percent of sessions on average, and that an external bug bounty found no universal jailbreaks across more than 1,000 hours of testing. Mythos 5 runs on the same underlying model with some safeguards lifted, replacing the Mythos Preview released in April.

Both models cost $10 per million input tokens and $50 per million output tokens. That’s double Opus 4.8’s rate, but, per Anthropic, less than half of what Mythos Preview ran. The price ladder is itself a product decision: capability is now a paid tier with policy gates attached.

Two other moves came with the launch. Anthropic imposed a mandatory 30-day data-retention policy on all traffic, overriding prior zero-retention enterprise agreements; the company says retained data won’t be used for training, only to defend against novel attacks and reduce false positives. CNBC reports Fable 5 scored more than 10 percent higher than Opus 4.8 on some benchmarks, and Stripe said the model migrated a 50-million-line Ruby codebase in a day, work it estimated at more than two months by hand.

Days earlier, Anthropic confidentially filed an IPO prospectus with the Securities and Exchange Commission, having recently closed a round at a $965 billion valuation. A safety-gated frontier launch, on the eve of going public, is its own kind of disclosure.

Sources