Claude Fable 5.1 arrives cheaper with fewer benign refusals

Anthropic released Claude Fable 5.1 on September 1, making the model generally available through its API and on AWS, Google Cloud, and Azure. A companion model, Claude Mythos 5.1, carries lighter safety filtering for vetted cybersecurity and life sciences work and stays restricted to trusted-access programs for US organizations.

Token prices are unchanged at $10 per million input and $50 per million output, but cache reads drop to $0.25 per million, a 75% cut. Anthropic estimates the net effect at roughly 25% lower cost for typical workloads and up to about 45% for heavily agentic use, where cached context drives most of the bill.

The other notable change is fewer refusals on benign prompts. Anthropic says its biology safety filters now trigger 85% less often on elementary biology and medical questions than at Fable 5's launch, and cybersecurity filters step in about 60% less per session. Fable 5.1 now permits defensive vulnerability identification while routing penetration testing, exploit generation, and binary analysis to Opus models.

On its own benchmarks, Anthropic reports Fable 5.1 scoring 55.8% on Terminal-Bench 4.0 for agentic coding, up from 42.0% for Fable 5, and 73.4% on CursorBench 3.2.0. Independent testing has not yet caught up. Early-access customers including Cognition and Jane Street said they were shifting coding traffic from Opus 5 to the new model.

For anyone building on the Claude API, the practical takeaway is a cheaper default for agent loops that reuse large context windows, plus a lower chance of a legitimate security or research query being blocked.