Claude Fable 5.1 and Mythos 5.1: Same Model, Different Safeguards

Anthropic just released Claude Fable 5.1 and Claude Mythos 5.1 — the same underlying model, but with different safety levels. Fable 5.1 is generally available; Mythos 5.1 is limited to trusted access programs for cybersecurity and life sciences.

Pricing and data retention

Fable 5.1 costs about 25% less than Fable 5 for typical token-billed workloads, thanks to reduced pricing on cache reads (inputs already processed and stored). For highly agentic work, savings can reach ~45%.

Enterprise Frontier Safeguards (EFS) give customers complete privacy — data stored in customer-controlled cloud infrastructure, not Anthropic's. EFS rolls out in phases starting fall 2025. Until then, eligible customers get zero data retention on Fable 5.1.

Benchmarks

Fable 5.1 shows gains across the board. Key numbers from the source:

• Terminal-Bench-Science 0.1: Fable 5.1 reaches ~50% accuracy vs ~25% for Fable 5, at similar cost.
• Terminal-Bench 4.0: At each effort level (low/med/high/xhigh/max), Fable 5.1 outperforms Fable 5. The gap between Fable and Mythos is expected to shrink with improved cyber safeguards.
• Humanity's Last Exam: Fable 5.1 with tools shows ~65% pass rate at max effort, before hitting the cost ceiling.
• CursorBench 3.2.0: Fable 5.1 scores ~72% at high effort, vs ~65% for Fable 5.

Fable 5.1 defaults to High effort in Claude Code, Medium in Claude Cowork and on Claude.ai. Lower effort settings yield comparable results to Fable 5 at lower cost.

Safeguard improvements

False positives (benign content flagged) drop 60% in cybersecurity. The model can now discover software vulnerabilities but not develop exploits. The biology side has an access program built with the US government — scientist enrollment opens soon.

Real-world impact

Investment firm Millennium reported Fable 5.1 found the root cause of a rare crash in their internal systems that engineers and other models couldn't explain after years of tries.