Anthropic shipped Claude Fable 5.1 on September 1, and the numbers around it read less like a routine point release than a repricing of the entire agentic workload market. AutomationBench, which scores models on end-to-end business tasks, moved from 17.1% on Fable 5 to 31.4% on Fable 5.1, an 84% jump inside a single decimal bump. Cache reads dropped 75%, from $1.00 to $0.25 per million tokens. Three months separate the two releases.

That combination is the story. Capability gains without a cost cut are a benchmark headline. Cost cuts without capability gains are a margin play. Both together, in the same release, are what actually reshuffles who runs which model in production.

The AutomationBench figure is the one worth staring at. At 31.4%, Fable 5.1 clears Opus 5's 26.9% on the same benchmark, meaning Anthropic's cheaper, faster tier now outperforms its flagship at the exact task category most relevant to outreach automation, lead research, and content pipelines. Terminal-Bench-Science 0.1 tells a similar story: 52.6% for Fable 5.1 against 24.7% for Fable 5, 29.0% for Opus 5, and 22.4% for GPT-5.6 Sol. Standard errors of 3.5 to 4.5 points per model mean the ordering at the top is real, not noise.

Base pricing didn't move. Input stays at $10 per million tokens, output at $50. The 1M-token context window and 128K max output are unchanged, and the knowledge cutoff still sits in June 2026. What changed is the cache-read multiplier, now 2.5% of base input on Fable 5.1 versus 10% on every other Claude model. Anthropic frames the real-world impact as roughly 25% lower effective cost on typical workloads and up to 45% on agentic ones, where cache reuse dominates spend.

The rest of the pricing sheet reads the same as before. Five-minute cache writes at $12.50, one-hour writes at $20, batch input and output at half of base, a 1.1x multiplier for U.S.-only inference, and web search at $10 per 1,000 searches on top of tokens. GDPval-AA v2 lands at 1,853 for Fable 5.1, 1,824 for Opus 5, 1,723 for Fable 5.

The reviewer at The New Stack noted that on real work, Fable 5 and 5.1 were hard to tell apart. That's a familiar pattern from the Mythos 5 enterprise-security rollout and the earlier UK AISI evaluation of Mythos 5: benchmarks move, day-one qualitative differences are subtle, the agentic gap widens quietly over weeks.

For teams routing outreach or content agents through the Claude API, the decision this month isn't whether Fable 5.1 feels smarter in a chat window. It's whether the 84% AutomationBench delta and the 75% cache-read cut change the unit economics of anything already in production. In most cases, they do.

Sources

https://venturebeat.com/technology/anthropics-claude-fable-5-1-and-mythos-5-1-arrive-with-a-75-cost-reduction-for-fable-cache-reads
https://techcrunch.com/2026/09/01/anthropics-new-fable-release-is-cheaper-less-restrictive/
https://thenewstack.io/claude-fable-upgrade-tested/
https://www.marktechpost.com/2026/09/01/anthropic-releases-claude-fable-5-1-and-claude-mythos-5-1-52-6-on-terminal-bench-science-and-75-cheaper-cache-reads/
https://datasciencedojo.com/blog/claude-fable-5-1-performance-and-safety/