On every paid Claude plan, Claude Sonnet 5 gives you a 1M token context window in the Claude app, Claude Opus 4.8, 4.7, 4.6 and Sonnet 4.6 give you 500K tokens, and outside of those models the window is 200K. The number is set by the model you pick, not by how much you pay: Pro, Max, Team and Enterprise all get the identical figure.
That single fact resolves the question most people are actually asking. If your chat keeps hitting a wall and you are wondering whether upgrading from Pro to Max fixes it, the answer is no. Max buys you more usage, not a longer conversation. What changes the ceiling is which model you select and how you structure what you send.
What a context window actually is
The context window is everything Claude can see while writing one reply. Anthropic's own documentation calls it a working memory: it holds the conversation history plus the new output Claude generates, and it is measured in tokens rather than words. A token is a small chunk of text; Anthropic's glossary puts it at roughly 3.5 English characters, and its help centre describes a 200K API window as about 500 pages of text or more.
Two things follow from that definition. First, the window is per conversation, not per account. Open a new chat and the window resets to empty. Second, the window is not storage. Nothing in it survives the conversation it lives in. Claude is not saving your uploaded PDF for later; it is reading it fresh, inside that one thread, and then it is gone.
The numbers, by plan and by model
Anthropic publishes context window figures per model, and applies them uniformly across paid tiers. Here is what the help centre currently states for chatting with Claude in the app.
- Claude Sonnet 5: a 1M token context window on all paid plans.
- Claude Opus 4.8, Opus 4.7, Opus 4.6 and Sonnet 4.6: a 500K token context window on all paid plans.
- Outside of those models, 200K tokens. Anthropic names Fable 5 only for Claude Code and the API, and publishes no figure for it in the chat app.
- Anthropic does not publish a separate figure for the free plan, so treat any number you see quoted for free accounts as unverified.
The paid tiers themselves, as listed on Anthropic's pricing page, are Free at $0, Pro at $20 per month billed monthly or $17 per month billed annually, Max starting at $100 per month with a choice of 5x or 20x more usage than Pro, Team at $25 per standard seat per month or $20 billed annually, and Enterprise at $20 per seat per month plus usage billed at API rates. Not one of those tiers is described as buying a larger context window.
The app, Claude Code and the API differ
This is where most confusion starts, because the same model can carry a different figure depending on where you use it. In Claude Code on a Pro, Max, Team or Enterprise plan, Sonnet 5, Fable 5, Opus 4.8, Opus 4.7, Opus 4.6 and Sonnet 4.6 all support a 1M token window, which is higher than the 500K those Opus models get in the chat app. Anthropic notes that Pro users need to enable usage credits to reach the 1M window for Opus models in Claude Code, and that Sonnet 4.6 requires usage credits on every paid plan except usage-based Enterprise.
The developer API is different again. Anthropic states that the Claude API can ingest 1M tokens with Sonnet 5, Fable 5, Opus 4.8, Opus 4.7, Opus 4.6 or Sonnet 4.6, and 200K or more with all other models. So if you read a blog post quoting a 1M window for Opus and then find your own chat capping out sooner, the writer was probably quoting an API figure at you. Same model name, different surface, different number.
| Where you are using Claude | Sonnet 5 | Opus 4.8 / 4.7 / 4.6 and Sonnet 4.6 | Other models |
|---|---|---|---|
| Claude app, any paid plan | 1M tokens | 500K tokens | 200K tokens; Anthropic does not publish an app figure for Fable 5 |
| Claude Code, paid plan | 1M tokens | 1M tokens (usage credits needed: Pro for Opus, all paid plans except usage-based Enterprise for Sonnet 4.6) | Fable 5: 1M tokens; others not listed |
| Claude API | 1M tokens | 1M tokens | Fable 5: 1M tokens; all others 200K or more |
In the Claude app, the tier you pay for does not appear anywhere in these numbers. A Max subscriber and a Pro subscriber running the same model in the same app get the same context window. What Max buys, in Anthropic's own words, is 5x or 20x more usage than Pro, not a longer memory inside a single chat.
Why the advertised number is not the whole story
A 500K figure does not mean you get 500K tokens of your own writing. Anthropic's documentation is explicit that everything in the request counts: the system prompt, every message in the thread, tool results, images, documents, and the output Claude generates for that turn, including its extended thinking. The advertised number is the total capacity of the room, not the space left for your furniture.
- The system prompt. Claude arrives at every conversation with instructions already loaded. You never see them, but they occupy the window.
- Attached files. A PDF, a spreadsheet or a screenshot is converted to tokens and sits in the window for the rest of the thread, re-read on every single turn.
- Every prior turn. Turn 40 carries the full weight of turns 1 through 39. This is why long chats feel fine and then suddenly do not.
- Claude's own replies. Its answers accumulate alongside your questions.
- Room reserved for the answer. Anthropic notes that a portion of the context window is always reserved for Claude's response, so the longest possible conversation is slightly smaller than the full window.
There is also a quality effect that nobody advertises. Anthropic's developer documentation names it directly: as token count grows, accuracy and recall degrade, a phenomenon it calls context rot. More context is not automatically better. A tidy 30K token conversation will often produce sharper answers than a bloated 400K one, which is a genuinely useful thing to know before you dump six documents into a chat and expect precision.
What actually happens when you hit the ceiling
In the Claude app, you usually do not slam into a wall. Anthropic states that when your conversation approaches the context window limit, Claude summarises earlier messages so the conversation can continue without a hard stop, and that you may see Claude organising its thoughts while it does. This automatic context management is available to paid plan users with code execution enabled, and the summarising itself does not count toward your usage limits.
That summarising is a compromise, not a rescue. Earlier detail gets compressed into a shorter recap, which means the exact figure you pasted in twenty turns ago may survive only as a paraphrase. If precision matters, do not rely on it.
You can still hit a hard stop. Anthropic's troubleshooting page documents the message: "Your message will exceed the length limit for this chat. Try attaching fewer or smaller files or starting a new conversation." That fires when a single message plus everything already in the thread will not fit. The suggested fixes are to break content into smaller chunks, summarise or extract the key sections first, or start a new conversation.
One more consequence worth knowing: long conversations cost you more, not less. Anthropic states that longer conversations which trigger automatic context management consume more of your usage limit. So the marathon thread you have been nursing for a week is quietly the most expensive way to use your subscription.
Practical ways to work inside the window
The fix is almost never a higher tier. It is better hygiene about what goes into the window in the first place.
- Start a new chat when the topic changes. Anthropic recommends starting a new conversation if you are approaching your usage limit in a longer chat.
- Pick the model with the window you need. In the app, Sonnet 5 carries 1M tokens where Opus 4.8 carries 500K. Model choice is the lever, not plan choice.
- Use Projects for reference material. Anthropic notes that Projects use retrieval-augmented generation, letting Claude work with larger amounts of information more efficiently, and that documents uploaded to a project are cached so only new or uncached portions count against your limits.
- Attach the relevant pages, not the whole file. Extract the section you actually need before uploading.
- Batch related questions into one message rather than spreading them across many turns.
Before you paste a long document into a chat, ask yourself whether you need Claude to read all of it or only to answer one question from it. If it is the second, paste the two relevant paragraphs. You will get a faster, sharper answer, you will burn less of your usage allowance, and you will push the length limit much further away.
A bigger context window is not memory
Here is the part most articles on this topic get wrong. They treat the context window as though it were Claude's memory of you, so a bigger number sounds like a better relationship. It is not that at all. The window is scoped to one conversation and it empties the moment you open a new one. A 1M token window and a 200K token window are equally forgetful across chats.
Buying a higher tier changes nothing about this. Max does not make Claude recall the contract you uploaded last Tuesday, because that contract was never stored as something Claude carries with it. It was tokens in a thread, and the thread ended. This is why people upgrade, feel briefly satisfied by the extra usage, and then find themselves re-uploading the same three documents a fortnight later.
The honest framing is that a context window is a reading capacity and a memory layer is a storage system. They solve different problems. If the thing frustrating you is having to re-explain your own material every time you start a chat, no context window figure on any plan is going to fix it.
That gap is what MemX is built for. You save the documents, photos, voice notes and receipts you keep needing, MemX reads them with OCR and transcription, and you ask a plain question later and get the answer plus the source document it came from. It is a place for the things you would otherwise re-upload, kept private by architecture with per-user isolation, customer-managed encryption keys and on-device capture. It does not extend anyone's context window. It just means you stop paying that window to re-read the same PDF every week.
01Does Claude Max have a bigger context window than Pro?
No. Anthropic publishes context window sizes per model and applies them to all paid plans equally. Max buys 5x or 20x more usage than Pro, not a longer conversation. Switching from Opus to Sonnet 5 in the app raises the window from 500K to 1M tokens; switching from Pro to Max does not.
02How many tokens is Claude's context window in 2026?
In the Claude app on any paid plan: 1M tokens for Sonnet 5, 500K for Opus 4.8, 4.7, 4.6 and Sonnet 4.6, and 200K outside those models. On the Claude API, those same newer models reach 1M tokens. Anthropic does not publish a separate free plan figure.
03Why does my Claude chat say it is too long?
Your latest message plus everything already in the thread no longer fits. Anthropic's message reads: "Your message will exceed the length limit for this chat. Try attaching fewer or smaller files or starting a new conversation." Attached files and every previous turn count toward the limit, not just what you just typed.
04Does Claude remember previous conversations?
The context window does not carry between chats. It covers one conversation and resets when you open a new one, whatever your plan tier or the model's window size. Claude does have separate memory features that work across chats, and a Project keeps uploaded documents available within that Project, but both are different mechanisms.
05Do attached files use up Claude's context window?
Yes. Anthropic's documentation states that everything in the request counts toward the window, including documents and images. An attached file is converted to tokens and re-read on every turn of that conversation, so a large upload early on shortens how far the chat can run afterwards.
