Anthropic ships Fable 5.1 and Mythos 5.1, cheaper and stronger on science

Anthropic released Fable 5.1 and Mythos 5.1, a point update to its flagship line that pairs lower prices with a real jump in scientific and agentic work. The two share the same underlying model and differ in access: Fable 5.1 is generally available, while Mythos 5.1 keeps tighter guardrails and is limited to vetted cyberdefenders and life scientists. The headline number is on science. On Terminal-Bench-Science 0.1, Fable 5.1 scores 52.6% against 24.7% for Fable 5, roughly doubling the previous result. On agentic coding, measured by Terminal-Bench 4.0, Fable 5.1 reaches 55.8% and Mythos 5.1 reaches 60.9%.

The pricing change is aimed squarely at people who run these models in loops. Cache reads drop 75% to $0.25 per million tokens, which Anthropic says works out to about 25% cheaper for typical workloads and up to 45% cheaper for agentic tasks. Input stays at $10 and output at $50 per million tokens, so the savings come from caching and repeated context rather than the raw per-token rate.

On the safety side, Anthropic says it cut false positives in cybersecurity use by 60%, so the model refuses legitimate defensive work less often, and added new protections against distillation. The company frames Mythos 5.1 as the version that can now find software vulnerabilities defensively, which is why access stays gated. Access is currently US-only, with international expansion planned.

Why it matters

If you build agents or run long coding and research loops on Claude, the cheaper cache reads and the science gains are worth a direct test on your own workload, since the biggest savings show up in exactly the repeated-context patterns agents use. If you do defensive security or life-science work, the gated Mythos 5.1 is the version to request, but plan around US-only access for now.