Anthropic Launches Claude Fable 5.1 with Coding Gains and 75% Cache Price Cut
The release pairs a broad performance jump in agentic and coding work with a 75% cut to cache pricing, the cost that scales with long-running agent workloads.
Reporting from 1 source: ASCII.jp.
Anthropic announced Claude Fable 5.1 and Claude Mythos 5.1 on September 1. Fable 5.1, now generally available, targets coding, knowledge work, and long-running agent tasks. Anthropic's tests show it more than doubled its predecessor's science benchmark score and topped its own Opus 5 and OpenAI's GPT-5.6 Sol on a coding benchmark. Cache read pricing fell 75% to $0.25 per million tokens. Mythos 5.1, a safety-relaxed research variant, stays limited to approved organizations.
Anthropic held headline rates flat at $10 per million input tokens and $50 per million output tokens, and put the discount into cache reads, which dropped from $1 to $0.25 per million tokens. The company estimates typical workloads run about 25% cheaper than Fable 5, and long-running agent tasks up to 45% cheaper.
Third-party testing is more mixed. Artificial Analysis scores high-load Fable 5.1 at 66 on its Intelligence Index against 61 for GPT-5.6 Sol, but at moderate settings the OpenAI model is cheaper and generates output faster.
Mythos 5.1 shares Fable 5.1's base model with relaxed safety restrictions for research, and access remains limited to vetted organizations in cybersecurity and life sciences.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.