The Claude Fable 5.1 context window is 1 million tokens — and unlike most million-token headlines, this one comes with pricing that makes the full window usable: $10 per million input tokens, $50 per million output tokens, and cache reads at just $0.25 per million, per the Claude Code changelog entry that made claude-fable-5-1 the default Fable model. Released on 1 September 2026, Fable 5.1 pairs that 1M window with a 128K maximum output, which means you can hand the model an entire codebase, a full site's content or months of business context in a single request and still get long-form work back out.
📺 Watch: NEW Claude Fable & Mythos 5.1 is ABSURD!
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside · Want AI SEO help 1-on-1? Book a free SEO strategy session →
Claude Fable 5.1 Context Window: The Exact Numbers
Here are the specifications that matter, drawn from Anthropic's published model details and the Claude Code changelog (version 2.1.257, which introduced Fable 5.1 as the default):
| Specification | Claude Fable 5.1 |
|---|---|
| Context window | 1,000,000 tokens |
| Maximum output | 128,000 tokens |
| Input price | $10 per million tokens |
| Output price | $50 per million tokens |
| Cache read price | $0.25 per million tokens |
| Cache write price | $12.50 (5-minute) / $20 (1-hour) per million tokens |
| Model ID | claude-fable-5-1 |
| Release date | 1 September 2026 |
To put the window in physical terms: a million tokens is roughly 750,000 words of English text. That is several full-length novels, a large production codebase, or every article on a mid-sized website — in context at once, not retrieved in fragments. The 128K output ceiling matters just as much for practical work, because a huge input window is only half the story if the model can respond with a few pages at a time.
If you want to turn long-context models like this into actual revenue — agents that hold your whole business in memory and execute daily — check out the AI Profit Boardroom → get the full playbook inside. Prefer 1-on-1 guidance on your AI SEO? Book a free SEO strategy session.
What Filling 1 Million Tokens Actually Costs
The economics are where Fable 5.1 gets genuinely interesting, because the raw window size is only useful if you can afford to fill it repeatedly.
Filling the entire 1M window with fresh input costs about $10 per request at the published $10 per million rate. Do that on every call of an agent session and costs stack up fast — but that is exactly what prompt caching exists to prevent. Cache reads on Fable 5.1 are priced at $0.25 per million tokens, which is 2.5% of the input price. On the previous Claude Fable 5, cache reads cost $1.00 per million — so Fable 5.1 cuts the price of re-reading cached context by 75%, per Anthropic's published pricing.
In practice, that changes how you architect long-context work. An agent that loads a large, stable knowledge base — your codebase, your brand guidelines, your client history — pays the full input rate once when the cache is written ($12.50 or $20 per million depending on cache duration), then re-reads that context for pennies on every subsequent turn. A full 1M-token cached context costs roughly $0.25 to read back per call instead of $10. Long agent sessions over big context went from a luxury to a rounding error, and if you are optimising spend in that direction, the guide on reducing Claude Code token usage covers the levers that compound with these cache prices.
The Context Window Inside Claude Code
The reason this spec matters right now is distribution: per the Claude Code changelog, Claude Fable 5.1 is the default Fable model in Claude Code as of version 2.1.257, with the 1M window and this pricing attached. You do not opt into the big window with a special endpoint or an enterprise tier — it is what the tool reaches for out of the box.
For coding work, a 1M window changes the day-to-day experience in three concrete ways:
- Whole-repository context. Medium and large codebases fit in context in one piece, so the model reasons over real cross-file relationships instead of a retrieval system's best guesses about which files matter.
- Longer sessions before compaction. Agent harnesses summarise or trim history as conversations grow; a window five times the size means far fewer of those lossy compressions in a working day.
- Room for tooling. MCP servers, skills and tool schemas all consume context before your task does. With a million tokens, a heavily connected setup — like the one described in the managed MCP servers guide — stops competing with your actual work for space.
The same logic applies beyond coding. If you run agent workflows that pull Claude Code into research or content systems — the pattern covered in the Hermes Agent with Claude Code write-up — the window is what lets an agent keep an entire project's state in view instead of paging it in and out. That is the architectural bet behind the Agent OS too: agents get dramatically more useful when context stops being the scarce resource.
📺 Watch: Claude Fable 5.1 Just Set a New AI Performance Record
How Fable 5.1's Window Compares
Context window size has become one of the main dividing lines between frontier models, so here is the honest framing rather than a spec-sheet victory lap.
Against its own family: Fable 5.1 keeps the same $10 input and $50 output pricing as Claude Fable 5 while cutting cache reads by 75%, so the practical cost of using the window heavily dropped even where headline prices held. The Claude Opus 5 vs Fable 5 comparison covers how the Fable line relates to Opus if you are choosing within the Claude lineup, and Anthropic has committed to keeping Fable 5.1 available until at least 1 September 2027, which matters if you are building systems you have to maintain.
Against rivals: raw window size is converging across the frontier, and a bigger number does not automatically mean better recall across that window — models differ in how reliably they use the middle of a very long context. The GPT-5.6 vs Fable 5 comparison goes deeper on the trade-offs, and the Goldie Bench write-up covers how these model brains compare in hands-on tests rather than on spec sheets. The short version: treat the 1M window as capacity you architect around, not a guarantee that every token in it gets equal attention.
One more family note: Fable 5.1 shipped alongside Claude Mythos 5.1, an invitation-only sibling model, but for anything you can actually sign up and build on today, Fable 5.1 is the relevant release.
Claude Fable 5.1 Context Window FAQs
How big is the Claude Fable 5.1 context window?
1 million tokens, with a maximum output of 128,000 tokens per response, per Anthropic's published model specifications following the 1 September 2026 release.
Is the 1M context window available in Claude Code?
Yes. The Claude Code changelog for version 2.1.257 lists Claude Fable 5.1 — with 1M context — as the new default Fable model, so the full window is available in the standard tool rather than gated behind a separate tier.
What does it cost to use the full context window?
Roughly $10 to fill the window with fresh input at $10 per million tokens. With prompt caching, re-reading that same context costs about $0.25 per million tokens — a 75% reduction in cache-read pricing compared with Claude Fable 5's $1.00 per million.
Is a bigger context window always better?
Not automatically. Long-context recall quality varies between models and across positions within the window, and a full window costs real money on uncached calls. The window is best treated as headroom for caching-heavy agent architectures, not as an invitation to stuff every request to the brim.
How does this compare with free ways to run Claude Code?
Pricing above applies to the paid API. If budget is the constraint, the Claude Code free options guide covers the routes that cost nothing and what you give up — context window included — when you take them.
Verdict: The Window Is the Feature, the Cache Price Is the Story
A 1M-token context window makes a good headline, but the number that changes behaviour is $0.25 cache reads. Together they mean you can keep an entire business's context permanently loaded in front of a frontier model and pay cents to consult it, which is precisely the foundation agent systems have been waiting for. If your workflows still treat context as scarce — trimming, summarising, retrieving fragments — Claude Fable 5.1 is the signal that it is time to redesign them around abundance.
If you want agents that put this window to work — holding your whole business in context and shipping daily output — check out the AI Profit Boardroom → start building inside AIPB. Want a 1-on-1 look at your AI SEO setup first? Book a free SEO strategy session.
