Anthropic splits its flagship model into public Claude Fable 5.1 and gated Claude Mythos 5.1

Anthropic released two versions of its newest model on September 1, 2026. Claude Fable 5.1 is generally available. Claude Mythos 5.1 is the identical underlying model with looser safety guardrails, restricted to vetted partners through two new programs: one for cybersecurity defense, one for life-sciences research done in partnership with the US government.

Cheaper, with fewer false positives

Fable 5.1 cuts cache-read pricing 75%, to $0.25 per million tokens. Per Anthropic, that lowers typical-workload costs about 25% and highly agentic workloads as much as 45%. The company also says its updated cybersecurity safeguards produce 60% fewer false positives, and Claude Code sessions see about 60% fewer safeguard interventions per session. That addresses a standing complaint about safety-tuned models: a filter that flags legitimate work as often as misuse costs users time without making anyone safer.

What the gate buys

Mythos 5.1's Cyber and Life Sciences Verification Programs let vetted defenders and researchers get direct answers to questions Fable would hedge on, such as discovering vulnerabilities instead of describing them defensively. Anthropic is also rolling out "Enterprise Frontier Safeguards" starting this fall, letting eligible customers run the model on their own cloud infrastructure with data never touching Anthropic's servers. The company told TechCrunch: "Anthropic has never trained on enterprise data without explicit permission, and never will."

The Fable/Mythos split admits what most vendors leave implicit: the same model can be too restrictive for some legitimate users and not restrictive enough to hand to everyone. False positives carry a cost that never appears on an invoice. If your teams hit safeguard refusals on ordinary work, measure how often, and re-test on Fable 5.1.