SAN FRANCISCO, Sept. 4 — Anthropic has unveiled Claude Fable 5.1 and Claude Mythos 5.1, which the company describes as its most advanced models yet for coding and knowledge work. The release, which landed this week, also offers an early look at how AI might accelerate scientific discovery. The two models share the same underlying architecture but differ in their safeguards.
Fable 5.1 is generally available, while Mythos 5.1 is restricted to trusted access programs, with protections designed specifically for cybersecurity and life sciences work. Anthropic says Fable 5.1 can now discover software vulnerabilities, though it cannot develop exploits.
Cost and Privacy Advances
For typical token-billed workloads, Fable 5.1 costs an estimated 25% less than its predecessor, Fable 5, thanks to reduced pricing on cache reads. On highly agentic work, the savings can reach roughly 45%. One early-access partner said they moved their Opus 5 traffic in Devin to Fable 5.1 on launch day, citing the lower cost per task and the new cache read pricing.
Anthropic also introduced Enterprise Frontier Safeguards, or EFS, which give customers complete privacy by storing data in customer-controlled cloud infrastructure rather than on Anthropic’s own systems. The company describes this as equivalent to zero data retention.
EFS will roll out to enterprise customers in phases beginning later this fall; until then, eligible customers can use Fable 5.1 with zero data retention.
Benchmarks and Real-World Testing
The company says safeguards have been improved to reduce false positives, with cybersecurity protections now blocking 60% fewer false positives than before. On benchmarks, Fable 5.1 outperforms Fable 5 across Terminal-Bench-Science 0.1, Terminal-Bench 4.0, Humanity’s Last Exam, and CursorBench 3.2.0 at various effort levels. Fable 5.1 defaults to High effort in Claude Code and Medium in Claude Cowork and on Claude.ai.
In testing by the investment firm Millennium, Fable 5.1 identified the cause of a rare crash — occurring in about one in a million runs — in internal systems that no engineers or other models had explained in four to five years. Early-access partners also reported qualitative gains, including better readability over long multi-step tasks, state-of-the-art trading intuition, and the ability to build a complex prototype in about three days.
Scientific Access
In biology, an access program developed with the US government will enable scientists to use Mythos 5.1’s advanced biology capabilities. Enrollment for researchers is expected to open soon.
