Anthropic has launched Claude Fable 5, which the company calls "the most capable model we've ever released to the public"1. Claude Fable 5 is a Mythos-class model equipped with safeguards designed to make it suitable for general use.
The release follows Anthropic's decision not to broadly deploy its predecessor, Claude Mythos preview, after internal training and testing revealed the model was "finding thousands of cybersecurity vulnerabilities". Anthropic determined that a model capable of identifying such flaws could also be used to exploit them, and instead provided Claude Mythos preview to defenders of critical software infrastructure to help fix vulnerabilities before they could be weaponized.
"It was the right call for the moment, but it was never the goal. We believe powerful AI should be safe and accessible. That's why we went to work on Claude Fable 5," a company speaker said during the launch.
Requests touching high-risk domains — specifically cybersecurity and biology — are automatically reviewed and then redirected to Claude Opus 4.8. Anthropic frames this as an intentional design choice: users retain access to a capable model's general functionality while the cyber and biology risks associated with Mythos-class capability are mitigated by routing those queries to a less capable system. The company acknowledged the safeguards "are broad today" but said it will continue refining them to better permit safe requests.
ANALYSIS The redirect-to-Opus-4.8 mechanism represents a tiered-capability safety approach: rather than uniformly restricting the model or withholding it entirely, Anthropic segments risk by domain and routes sensitive queries to a model with a lower capability ceiling.
On the capability side, Anthropic says Claude Fable 5 "can stay with a problem far longer than any model before it" and can operate for days without intervention. The company positions it for use in finance, research, economics, law, and other complex tasks that previously required constant human supervision.
Separately, a report indicates Anthropic has "significantly adjusted access permissions" for Claude Fable 5, with some users losing access rights as the company accelerates expansion2. The report's headline references the permission changes but provides limited additional detail on the scope or rationale of the adjustments.
ANALYSIS The access-permission changes, occurring alongside a launch framed around broad public availability, suggest Anthropic is actively managing the rollout — potentially gating access by user tier or geography even as it markets Claude Fable 5 as its most widely available frontier model.
The launch arrives as Anthropic faces competitive pricing pressure. Chinese open-weight alternatives such as Zhipu's GLM-5.2 have seen surging adoption on platforms like Vercel, operating at roughly one-fifth the cost of Claude Opus 4.8 ctx.
ANALYSIS By tying Claude Fable 5's safety infrastructure to Claude Opus 4.8, Anthropic effectively keeps Opus 4.8 operationally central even as its standalone commercial position faces pressure from lower-cost competitors.