Claude Fable 5.1 Context Window: The Full 1M Guide

Wondering how big the Claude Fable 5.1 context window really is and what you can do with it? The quick version is just below: it’s a 1 million token context window by default, paired with 128k max output tokens — confirmed in Anthropic’s official platform release notes for the 1 September 2026 launch. I’ve been running Fable 5.1 on long SEO and automation jobs since it became the default model, and in this guide I’ll break down what the numbers mean, what the pricing looks like, and how to actually exploit a context window this large without burning money.

Claude Fable 5.1 Context Window: The Official Numbers

Let’s start with the facts as Anthropic published them. On 1 September 2026, Anthropic released Claude Fable 5.1 (claude-fable-5-1) alongside Claude Mythos 5.1, and the platform release notes state plainly: “Both models support a 1M token context window by default, 128k max output tokens.” Two details in that sentence deserve emphasis.

First, “by default”. Previous million-token contexts in the Claude line-up were beta features you enabled with special headers, often gated by usage tier. On Fable 5.1 the 1M window is just… the window. Every request gets it. That changes it from a party trick into infrastructure you can build on.

Second, 128k max output tokens. A giant input window is only half the story — if the model can only answer with a few thousand tokens, long-document work still means stitching. 128k of output means Fable 5.1 can read a huge corpus and produce a genuinely long deliverable in one shot: a full content cluster, a complete refactor, a whole documentation set.

For scale: 1 million tokens is roughly 750,000 words of input. That’s an entire mid-sized website’s content, or every blog post most businesses have ever published, in a single prompt.

What It Costs to Fill 1M Tokens (and the Cache Trick)

Here’s the part most coverage skips. Base pricing is $10 input / $50 output per million tokens — the same as Claude Fable 5. So yes, stuffing the window costs about $10 per fully-loaded request. Do that naively in a loop and your bill gets silly fast.

The release notes bury the real headline: “cache reads cut to $0.25 per MTok”. Anthropic spells it out — prompt cache reads on Fable 5.1 and Mythos 5.1 cost 0.025x the base input price, compared with 0.1x on other models, with cache write prices unchanged. In practice that means the smart pattern is: load your big corpus once, cache it, then run dozens of questions against the cached context at $0.25 per million tokens instead of $10. That’s a 40x discount on every repeat read, and it’s the single most important fact about the Claude Fable 5.1 context window if you’re paying the bill.

Adaptive Thinking, Effort Controls and the Gotchas

A few more release-note details that affect how you use all this room:

• Always-on adaptive thinking. Fable 5.1 decides how hard to think per request, and you cannot switch thinking off — thinking: {"type": "disabled"} isn’t supported. Budget for thinking tokens in long jobs.
• Per-message effort changes (beta). Since 10 September 2026 you can change effort mid-conversation on Fable 5.1, Mythos 5.1 and Opus 5 via the mid-conversation-output-config-2026-07-01 beta header — crank effort up for the hard step, drop it for boilerplate.
• tool_choice restriction. The any and tool options for tool_choice return a 400 error on Fable 5.1; only auto and none work. If your agent framework forces tool calls, adjust it.
• Availability. Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud (Vertex AI) and Microsoft Foundry — plus it’s the default Fable model in Claude Code since v2.1.257 (1 September 2026, per the Claude Code changelog).

I dig into the day-to-day feel of the model in my full Claude Fable 5.1 review, and if you’re tuning how hard the model works per task, my Claude Code effort level guide pairs naturally with the new per-message effort beta.

How I Use the Claude Fable 5.1 Context Window for SEO

Theory is nice; here’s what a million tokens is actually good for in an SEO business:

Job	Old way (small context)	With the 1M window
Whole-site content audit	Chunk pages, summarise, merge summaries, lose nuance	Load every page at once; ask for cannibalisation, gaps and internal-link fixes directly
Topical authority mapping	Sample your top pages	Feed the entire site plus competitors’ sitemaps in one prompt
Content refresh sprints	One article at a time	Load 50 dated posts, get a prioritised rewrite plan with 128k tokens of output
Agent memory	Aggressive summarisation between steps	Keep the whole job history in context; cache it at $0.25/MTok per read

The workflow that’s paid for itself already: dump the full text of a client site into one cached prompt, then interrogate it — “which pages target the same query?”, “which posts contradict each other?”, “where are the internal-link orphans?”. Answers come back grounded in the whole site rather than a sample, and every follow-up question rides the $0.25 cache-read rate. That’s the kind of leverage we template out inside the AI Profit Boardroom so members can run it on their own sites the same afternoon.

The bottom line on the Claude Fable 5.1 context window

The Claude Fable 5.1 context window is 1 million tokens by default with 128k output tokens, at $10/$50 per MTok and — the sleeper feature — $0.25 per MTok cache reads, all per Anthropic’s official release notes of 1 September 2026. The window itself isn’t the innovation; making it default and making repeat reads 40x cheaper is. If you do SEO, that combination turns whole-site analysis from an engineering project into a single well-structured prompt. Just remember the constraints: thinking is always on, tool_choice forcing is gone, and an uncached full window costs real money per call.

FAQ: claude fable 5.1 context window

How big is the Claude Fable 5.1 context window?

1 million tokens by default, per Anthropic’s platform release notes (1 September 2026). Both Fable 5.1 and Mythos 5.1 get the 1M window with 128k max output tokens.

Do I need a beta header for the 1M context window?

No. The release notes say both models support the 1M token context window by default — it’s standard, not an opt-in beta.

How many output tokens can Claude Fable 5.1 produce?

Up to 128k output tokens per response, per the official release notes — enough for genuinely long deliverables in a single call.

What does it cost to use the full context window?

Base pricing is $10 input / $50 output per million tokens (unchanged from Fable 5). The key saving: prompt cache reads cost $0.25 per MTok — 0.025x base input — so cache your big context and interrogate it cheaply.

Can I turn off thinking on Claude Fable 5.1?

No — adaptive thinking is always on, and thinking: {"type": "disabled"} isn’t supported on Fable 5.1 or Mythos 5.1.

Where can I use Claude Fable 5.1 today?

Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud (Vertex AI) and Microsoft Foundry — and it’s the default Fable model in Claude Code as of v2.1.257.

Related reading

• Claude Fable 5.1 Review: Worth the Upgrade?
• Claude Code Effort Level: Full Guide to /effort
• Claude Code Plugin Eval Command: Full 2026 Guide

Last updated September 2026. This is the living guide to claude fable 5.1 context window — it gets updated as the tools change.