Claude Fable 5.1 hits 55.8% on Terminal-Bench 4.0 as cache reads drop 75%
Anthropic shipped Fable 5.1 and the trusted-access Mythos 5.1, scoring 55.8% on Terminal-Bench 4.0 agentic coding against Fable 5's 42.0%, and 52.6% on Terminal-Bench-Science against 24.7%. Cache reads fall 75% to $0.25 per million tokens, cutting typical workloads about 25% and highly agentic runs up to 45% — a frontier capability jump arriving with a price cut rather than a premium. Mythos 5.1 designed protein binders with 10x the affinity of the best Adaptyv Bio competition entries at a hit rate near 50%, against the 10-15% typical in protein design today. Fable 5.1 can now be used for vulnerability discovery but not exploit development, with 60% fewer false-positive safeguard blocks in cybersecurity.
Source: anthropic.com ↗
Its hit rate (that is, the number of designs that were viable binders) was the strongest we've measured to date: it reached nearly 50% across 12 targets.
Why this matters
- → 55.8% on agentic coding benchmarks — 33% leap over Fable 5.
- → Cache pricing cuts typical costs 25%, agentic workloads up to 45%.
- → Protein design hit rate near 50% vs. 10–15% industry standard.