Anthropic has introduced Claude Fable 5.1 and Claude Mythos 5.1, which are essentially the same model with two different safeguard tiers. Fable 5.1 is broadly released; Mythos 5.1 is available only through trusted access programs, featuring safeguards specifically tailored for cybersecurity and life sciences work.

Fable 5.1 is expected to be roughly 25% cheaper than Fable 5 for standard token-based tasks, mainly driven by price cuts to cache reads, where the model reads previously processed and cached data. For agentic tasks (which rely heavily on context and tools), savings can reach up to 45%. Base list pricing remains identical to Fable 5 at $10 per million input tokens and $50 per million output tokens, with only cache read pricing reduced.

Data Retention and Enterprise Frontier Safeguards (EFS). The new EFS system provides enterprise clients with complete privacy (equivalent to a zero-data-retention policy) while maintaining state-of-the-art abuse prevention. The key difference: data is stored in cloud infrastructure controlled directly by the customer, not Anthropic. EFS will roll out in phases starting this fall, and until then, eligible customers can use Fable 5.1 with zero data retention.

Safeguards have also improved. Anthropic reduced cybersecurity false positives by 60%—resulting in fewer mistaken alerts on benign content. Notably, Fable 5.1 can now be used to identify software vulnerabilities, though not to build exploits for them. In biology, Anthropic is partnering with the U.S. government on a trusted access program to make Mythos 5.1’s advanced biological capabilities available, with enrollment for scientists expected to open soon.

Performance figures published by Anthropic:

  • Terminal-Bench-Science 0.1: 52.6% (Fable 5: 24.7%, Opus 5: 29.0%).

  • Terminal-Bench 4.0: 55.8% (Mythos 5.1: 60.9%), with the performance gap between versions reflecting tasks where legacy safeguards intervened; the gap is expected to narrow under the updated safeguards.

  • Humanity’s Last Exam: 60.9% without tools, 65.0% with tools.

  • CursorBench 3.2.0: 73.4%.

  • GDPval-AA v2: 1853.

ImageOne demonstration of Fable 5.1’s vulnerability-hunting capabilities occurred at hedge fund Millennium, where the model determined the cause of an extremely rare crash in an internal system that engineers (and all other models) had failed to explain for years; it disassembled a vendor library, cross-referenced it with a core dump, and tracked down the bug within the library.

As for Mythos 5.1, access is currently limited to vetted U.S. organizations (cyberdefenders and life scientists), and Anthropic is collaborating with the U.S. government to broaden access to domestic and international partners.