Claude Fable 5.1

Claude Fable 5.1 is an incremental release on top of Fable 5, out on 1 September 2026. Same 1M-token context window, same 128K output ceiling, same headline price. Anthropic put the work into long-running agentic coding, multistep research, and document, spreadsheet, and slide work.

What actually changed

The price per token didn’t move: $10 per million in, $50 per million out. The cache did. Reading from the prompt cache now costs $0.25 per million tokens. On every other Claude model a cache read costs 10% of the base input price; on Fable 5.1 it costs 2.5%. If you run long agent sessions over a fixed body of context, that’s where most of your bill sits, and it just fell by three quarters.

Three changes break code written against Fable 5. Forced tool use returns an error. Earlier models can no longer read its thinking blocks. And editing an earlier turn invalidates the thinking blocks that came after it. Anthropic also added per-message effort, turn-scoped system messages, and readable progress updates between tool calls, all in beta.

Where it sits in the lineup

Anthropic’s own guidance is to start with Opus and reach for Fable only when your evaluations on Opus at a high effort setting still fall short. That’s a fair description of the trade, and Claude Opus 5.5 made it sharper: Anthropic benchmarks it at Fable’s level on most work. Fable is slower and two and a half times the price, and that only pays off on a narrow band of work: hard reasoning, and agent runs long enough that a dropped detail halfway through costs you the whole session.

The line also has a history worth knowing about if you plan to depend on it. Fable 5 was pulled worldwide for nearly three weeks in June 2026 under a US government order before returning in July. I wrote about what that means for building on a single frontier model.

What it’s actually good at

Fable got a rough reception in the AI circles on social media when 5.1 landed. My experience ran the other way: until Opus 5.5 arrived it was my favorite model, and at work almost everything went through it. Not because it answers questions well, which would be a waste of it, but because it holds a hard problem in view long enough to be useful as a sparring partner, an architect, a designer, and a reviewer. That’s four different jobs, and it’s good at all of them.

Then Opus 5.5 arrived on 22 September, and as a snapshot the picture changed. Opus is now so capable and so efficient that it’s my default for almost everything. Fable adds a little edge on very big, long-running tasks, and that’s the band of work I keep it for. I expect this to flip again with the next Fable release.

The corollary is the same as it was for Fable 5: asking it simple things is a waste of money and quota. Route those elsewhere.

On cost, the API rate is the wrong number to look at if you work through the apps. A Team Premium seat is $100 a month and includes Fable rather than metering it on top, which changes the calculation completely. At API rates the same volume of work would be unthinkable; on a seat it’s just the subscription.

Claude Mythos 5.1

Anthropic shipped a sibling model, Mythos 5.1, on the same day. Same capabilities and same pricing, under a more permissive safety posture, available by invitation only through Anthropic’s Project Glasswing. There’s no public sign-up, so for almost everyone reading this it’s a name in a system card rather than a choice to make.

The Verdict

Best for: Very big, long-running agentic tasks where a little extra edge over Opus 5.5 is worth two and a half times the price.