Anthropic has announced Claude Fable 5, a new AI model described as the company’s first broadly available release from its more capable Mythos class. The launch is notable because Anthropic had previously kept Mythos-level systems away from general users over safety concerns, especially around advanced cybersecurity and sensitive scientific tasks.
According to The Verge, Fable 5 is positioned as Anthropic’s most powerful public model so far, with stronger performance in software engineering, knowledge work, and vision tasks. The company says the advantage becomes more visible as jobs get longer and more complex, which is exactly the kind of workload where frontier AI tools are now competing hardest.
Claude Fable 5 Adds Guardrails For Risky Requests
The key difference is not only raw capability. Claude Fable 5 ships with safeguards that can route high-risk prompts away from the Mythos-class model and back to Claude Opus 4.8. The Verge reports that Anthropic named cybersecurity and biology as two areas where requests may be blocked or redirected, while Axios says the routing can also apply to chemistry and model-distillation requests.
Anthropic says 95 percent of Fable sessions in testing stayed on Fable without falling back to Opus 4.8. That figure matters because an overly aggressive safety layer could make the new model feel inconsistent for normal users. Axios also reports that Anthropic ran internal safety tests, red-team work, and outside adversarial testing before release.
The pricing puts Fable 5 above Anthropic’s previous flagship tier. The Verge reports a rate of $10 per million input tokens and $50 per million output tokens for both Fable 5 and Mythos 5. That is higher than Claude Opus 4.8, but Anthropic argues the stronger model could still reduce total cost on complex tasks if it needs fewer prompts and less correction.
Anthropic is also introducing Claude Mythos 5, which uses the same underlying model but with fewer restrictions in some areas. Access appears limited to organizations already connected to the company’s Project Glasswing program, with a broader trusted-access system planned later.
One caveat: Anthropic’s public Claude model overview still listed Opus 4.8 as the top general model when checked, so developer availability and exact API naming may take time to settle. For now, Fable 5 looks like Anthropic’s attempt to bring Mythos-level intelligence to general users without opening the highest-risk capabilities by default.
You can follow more developments in Technowatt’s Artificial Intelligence coverage.