Anthropic released Claude Fable 5.1 and highlighted a 75% cut to cache read prices ($1 → $0.25 pe...

Anthropic released Claude Fable 5.1 and highlighted a 75% cut to cache read prices ( $1 →$ 0.25 per million tokens) right at the top of their announcement. They estimate this will reduce costs by 25% for typical workloads and up to 45% for highly agentic workloads.

What they do not mention is the increase in verbosity. Artificial Analysis data shows Fable 5.1 (max with fallback) costs $3.69 per Intelligence Index task, about 20% more than Fable 5 at$ 3.14. The model generates more output tokens: 45k versus 36k for Fable 5. That is a 25% increase in output tokens alone. This more than offsets the cache savings, and you end up paying more.

This fits the pattern we have seen with Anthropic’s higher-end models: strong intelligence paired with higher token usage. They still have a long way to go before they can catch up with OpenAI on token efficiency.

Cost per Intelligence Index task breakdown across token types — Weighted average cost per task on the Artificial Analysis Intelligence Index segmented by token type (Answer, Reasoning, Cache Write, Cache Hit, Input).

Output tokens per task comparison highlighting Fable 5 vs Fable 5.1 — Output tokens generated per task comparing Claude Fable 5 (36k tokens) against Claude Fable 5.1 (45k tokens). Source: Artificial Analysis.