Claude Fable 5.1 lands with a 45% price cut on agentic coding — and a zero-data-retention option for enterprise
Anthropic's Claude Fable 5.1, released 1 September 2026, more than doubles Fable 5's Terminal-Bench-Science score (24.7% to 52.6%), leads CursorBench 3.2 at 73.4% for agentic coding, and cuts costs by up to 45% on heavily agentic workloads — while adding a zero-data-retention option that removes a real barrier to enterprise AI-assisted development adoption.
14 September 2026
Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on 1 September 2026, and the coding-specific numbers are worth pausing on rather than filing under routine model-update noise. Fable 5.1 more than doubles Fable 5’s score on Terminal-Bench-Science, from 24.7% to 52.6%, and leads CursorBench 3.2 at 73.4% for agentic coding tasks — features that span an entire codebase, code review, performance work, and multi-day autonomous sessions rather than single-file completions. It matches Fable 5’s coding power using roughly half the tokens, and communicates in plainer language, which is the detail that actually reaches a non-technical buyer: faster turnaround on the same scope of work.
The cost curve is the real story
Fable 5.1 costs around 25% less than Fable 5 for typical workloads, and up to 45% less for heavily agentic tasks, driven largely by cache reads dropping 75% in price. That’s a meaningful shift in the unit economics behind AI-assisted development: the tools doing the most autonomous, multi-step work — the kind that used to be the most expensive to run — just got proportionally the cheapest. Anthropic also added a zero-data-retention option at launch, which removes a barrier that had genuinely kept large, compliance-conscious organisations from adopting the previous generation of Claude models for anything touching sensitive codebases or client data.
Why this matters beyond the benchmark chart
None of this changes what a founder or CTO actually needs from a build: a team that uses these tools with the same review discipline, testing rigour and architectural judgement it would apply anyway. What it does change is the economics and the compliance calculus behind AI-assisted development engagements. A model that’s simultaneously more capable at sustained, autonomous work and cheaper to run on exactly that work shifts what’s realistic to scope inside a fixed budget — and a genuine zero-data-retention option is the kind of detail that turns “we can’t use AI tooling on this project” into “we can, provided it’s configured correctly.”
So what
If your last conversation about commissioning an AI-assisted build stalled on cost or on data-handling concerns, both of those objections just got weaker. The gap that still matters is the same one it’s always been: whether a development partner treats a model like Fable 5.1 as a productivity multiplier inside proper engineering process, or as a shortcut around it. See our AI-assisted development approach for how we scope projects around tools like this, or get in touch if you want to know what a Fable 5.1-era build actually costs for your project.