Claude Fable 5.1 launches with a 1M-token context window and cache reads cut 75% — Anthropic bets on cheaper, longer agentic runs
Anthropic released Claude Fable 5.1 on 1 September 2026 with a 1M-token context window, 128K max output, and cached-token pricing cut by 75%, alongside Claude Mythos 5.1 — both models are the first from Anthropic to carry invisible watermarking in line with EU AI Act rules.
9 September 2026
Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1 on 1 September 2026, both carrying a 1M-token context window, 128K maximum output tokens, and adaptive thinking switched on by default. The change most relevant to anyone running AI coding agents at volume is pricing: cache-read costs dropped 75%, from $1.00 to $0.25 per million tokens, which Anthropic measures as roughly 25% cheaper on typical workloads and up to 45% cheaper on agentic ones — the workloads that repeatedly re-read the same codebase context on every turn. Both models are also the first Claude releases to carry invisible watermarking in text and generated files, a response to EU AI Act requirements that apply to models released after 2 August 2026.
Anthropic says Fable 5.1 is specifically better at handling software projects that span an entire application — reviewing code across files, troubleshooting problems that originate in external libraries, and running multistep technical work without losing track of earlier context. That’s a direct response to where long-running coding agents have struggled: a 1M-token window matters less for a single function than for a task that needs to hold an entire codebase’s structure in view across dozens of tool calls.
Why the pricing detail matters more than the benchmark
Every new frontier model launch gets covered as a benchmark story, but for teams actually running agentic coding tools day to day, the cache-pricing cut is the more consequential change. Agentic workflows re-send large chunks of context — file trees, prior conversation, tool outputs — on nearly every step, and cache reads are where that cost accumulates fastest. A 75% cut to that specific line item changes the economics of running coding agents on larger, longer-lived tasks rather than small, isolated ones, which is exactly the direction the market has been moving.
So what
If your AI-assisted development costs have been climbing as your team runs agents on bigger tasks, this is the kind of underlying platform shift worth revisiting your setup for — the tools and models change fast enough that a workflow priced out six months ago may look different today. See our AI-assisted development work, or get in touch to talk through what a properly engineered agentic workflow looks like for your team.