ChatGPT Context Window Full: What Happens and What to Do Next

Key Takeaways

• A "chatgpt context window full" message means one conversation has used up all the text ChatGPT can hold in view at once. It is a limit on that single chat, not on your account.
• Everything in the chat counts toward the limit: your messages, ChatGPT's replies, pasted documents, uploaded files, and tool output.
• Before the hard stop, long chats usually degrade quietly. ChatGPT starts forgetting or contradicting earlier messages because the oldest parts of the conversation have slipped out of view.
• When the window fills, the fix is a new chat. Ask ChatGPT for a short summary first, while you still can, and carry that summary into the fresh conversation.
• OpenAI's Help Center says that on the Free plan, ChatGPT can show either "Context window full" or "Chat memory full" when a conversation fills up, and that you can start a new chat right away.
• Upgrading to a paid plan gives you a larger context window, but no plan has an unlimited window. The habits in this guide matter on every plan.

What Is a Context Window?

A context window is the amount of text a language model can hold in view at one time. Think of it as ChatGPT's working memory for a single conversation. Every time you send a message, ChatGPT looks back over the conversation so far — your prompts, its replies, anything you pasted in, and the output of any tools it used — and generates its next reply from that whole picture.

That picture has a fixed size, measured in tokens. A token is a small chunk of text, roughly three-quarters of an English word on average, so a thousand tokens is roughly 750 words. The exact size of the window depends on the model and on your plan, and OpenAI changes these figures over time, so treat any number you read as a snapshot, not a permanent guarantee.

This is different from two things people often confuse it with. First, it is different from a usage or message limit, which caps how much you can send over a period of time. A context window caps how long one conversation can grow. Second, it is different from ChatGPT's saved Memory, which stores facts about you across chats. Memory and context are separate systems, and turning Memory off or deleting saved memories does not free up space in a full conversation.

It is also worth separating the ChatGPT product from the OpenAI API. The API exposes much larger context windows for some models — GPT-5-family models are documented with a 400,000-token class window, for example — but those API figures do not apply inside chatgpt.com or the ChatGPT apps. Inside ChatGPT, OpenAI publishes separate, smaller product limits that depend on the plan and on whether you are using an instant or a reasoning model.

What Happens When the Context Window Fills Up?

ChatGPT handles a full window in two ways, and most users meet the quiet one first.

The quiet version: the chat starts forgetting

Long before the hard stop, the oldest messages in a conversation begin dropping out of what the model can see. Nothing is deleted from your chat history — you can still scroll up and read every message — but ChatGPT itself can no longer see the top of the thread. That is why a long chat starts to drift: it contradicts something you agreed on an hour ago, asks for information you already gave it, reverts to an earlier version of a plan you already changed, or repeats questions it already asked.

This is expected behaviour, not a bug. The model can only reason about what is inside its window, and once the conversation is longer than the window, the earliest turns are simply no longer part of the picture. If your chat is getting vague, repetitive, or forgetful, the window is almost certainly the cause, even if no warning has appeared.

The loud version: the chat stops

Eventually the conversation reaches the point where ChatGPT cannot take any more input in that thread at all. OpenAI's Help Center describes the notices users see as "Context window full" and "Chat memory full", and says both refer to how much information ChatGPT can use in the current conversation. On the Free plan, either notice can appear when a conversation fills up.

In older versions of the product, users saw a message reading "You've reached the maximum length for this conversation, but you can keep talking by starting a new chat." The wording has changed over time, and different plans and models may show slightly different notices, but the meaning is the same: the input box is effectively locked for that thread, and the only way forward is a new chat.

One important detail, confirmed by OpenAI's own help pages as summarised in recent reporting, is that a full conversation has no reset time. You do not need to wait for an allowance to refill, and starting a new chat works right away, including on the Free plan. The original conversation stays in your history, readable but unusable for further messages.

How Big Is the Window on Each Plan?

OpenAI publishes the ChatGPT context window for each plan on its pricing page, and the figures have changed as new model generations have arrived. According to reporting that read the pricing page on 7 November 2026, the instant-model window was 27,000 tokens on the Free plan, 54,000 tokens on Go and Plus, and 128,000 tokens on Pro. For reasoning models, the same snapshot put Go and Plus at 256,000 tokens and Pro at 400,000 tokens.

Older snapshots show different numbers, because they describe different model generations — for example, a Help Center table for GPT-5.3 listed 16,000 tokens for Free instant use and 32,000 for Plus. That is not a contradiction so much as a reminder that the window belongs to the model generation in use, not to ChatGPT in general. Check the current pricing page before planning around any figure, and do not trust a number from a blog post without a date.

OpenAI also publishes approximate page counts for how much you can type or paste in at once. In the 7 November 2026 snapshot, Free instant use was described as about 12 pages, Go and Plus as about 40 pages, and Pro as about 250 pages. Those are approximations: system instructions, saved memories, and features in use all take up part of the window, so the space left for your own input is smaller than the total.

There is no published fixed character limit for a single message. What OpenAI publishes is the context window itself, plus those approximate page counts. If a single pasted block is very long, shortening it can help even when the conversation as a whole is short.

Why Your Chat Fills Up Faster Than You Expect

A few habits burn through context surprisingly quickly.

Long pasted documents. Pasting a whole report, a long transcript, or a large block of code counts every token. A few long exchanges can fill a chat, as OpenAI's Help Center puts it: your messages and ChatGPT's replies both take up space.

Long replies. ChatGPT's own output counts too. A chat where every reply is a 2,000-word essay fills up roughly twice as fast as one with short, focused answers. Asking for shorter replies is a genuine fix, not just a preference.

Files and tool output. Uploaded files, images, data-analysis results, and web-browsing output all share the same window. A data-analysis session with large tables can fill a chat in a handful of turns.

Repeating context. Re-pasting the same background, brief, or document in every message multiplies the cost. Paste it once, then refer to it by name.

One mega-chat for everything. The most common cause is simply using a single conversation as a permanent workspace for a whole project. That is exactly the use case the limit punishes. A chat that has been running for weeks, across dozens of topics, is almost certainly carrying a large amount of context that no longer matters.

What to Do When ChatGPT Says the Context Window Is Full

Do not panic, and do not delete anything. Your conversation is safe in your history. Work through these steps in order.

1. Ask for a summary while you still can

If the chat still accepts messages, ask ChatGPT to summarise the conversation so far: the key decisions, the facts it should remember, the open questions, and the current state of the work. Be specific about what matters — a generic summary wastes space on things you do not need. Copy that summary somewhere safe before you do anything else.

If the chat is already locked, you cannot ask for a summary inside it. You can still read the whole thread yourself and write a short summary by hand, or paste the important parts into a new chat and ask ChatGPT there to condense them.

2. Start a new chat and carry the summary in

Open a fresh conversation and paste the summary as your first message, followed by a clear statement of what you want to do next. The token count resets to zero in a new chat, so you start with a full window again. Keep the handover short: decisions, constraints, and current status, not the full history. Some users report that ChatGPT offers a "continue in a new chat" option with an automatic summary; that option has been reported by users rather than documented in the Help Center, so check whatever summary it produces before relying on it.

3. Put lasting facts somewhere durable

For information you will need across many chats, use ChatGPT's Memory or Custom Instructions for stable facts about you and your preferences, and use Projects for ongoing work that needs its own files and instructions. Neither of these stops a single thread from overflowing, but they reduce how much background you need to paste into every new chat. Remember that Memory capacity and a single chat's context window are different limits: a full memory is a separate problem with a separate fix.

4. Consider whether a bigger window would actually help

Moving from Free to Go or Plus doubles the instant-model window in the November 2026 snapshot, from 27,000 to 54,000 tokens, and Pro reaches 128,000 tokens for instant use and 400,000 for reasoning. If you regularly work with long documents, a larger window is a real improvement. But it is a bigger bucket, not an infinite one. A project that fills 27,000 tokens will eventually fill 54,000 too, and the summary-and-restart habit will still matter. Upgrade for the headroom, not as a substitute for good habits.

For more on how ChatGPT's allowances are changing, see our report on ChatGPT and Codex sharing one token budget in December. Usage limits and context limits are different things, and the December change concerns the usage side.

How to Stop Hitting the Limit

The goal is not to use less of ChatGPT. It is to spend your context on the work that matters and stop paying for context you no longer need.

Start a fresh chat per task, not per project. One conversation per deliverable — one article, one analysis, one debugging session — keeps each thread short and focused. When the task changes, start a new chat.

Summarise early, not at the crisis. When a chat passes the point where you would struggle to reconstruct it from memory, ask for a summary and save it. Doing this at 70 percent is far easier than doing it at 100 percent, when the chat may already be locked.

Ask for short replies by default. Tell ChatGPT to keep answers concise unless you ask for detail. Shorter replies mean slower context burn and, as a bonus, faster reading.

Paste less, reference more. Instead of pasting the same document into every chat, paste it once, ask for a condensed version, and carry the condensed version forward. For recurring work, a one-page brief beats a fifty-page source document in every new thread.

Split long documents before you paste them. If you need ChatGPT to work through a long report, split it into sections and work through them one at a time, carrying forward only the conclusions. Feeding the whole document in one message is the fastest way to fill a window.

Watch for the warning signs. Repetition, contradictions, forgotten instructions, and answers that ignore earlier decisions are all signs that the oldest part of the conversation has dropped out of view. Treat them as a signal to summarise and restart, not as a reason to re-paste everything and push on.

Keep a handover note outside ChatGPT. A short note in your own files — the goal, the decisions so far, the open items — is the most reliable way to restart a conversation. It does not depend on ChatGPT being able to summarise itself, and it works even after a chat is locked.

If you are weighing whether to stay with ChatGPT or try another tool for long-document work, our guide to ChatGPT alternatives worth trying and when to switch compares the options, and our roundup of free AI tools in 2026 covers what you can use without paying.

Context Window Full vs. Other ChatGPT Limits

Because several different limits can look alike, it helps to tell them apart.

A context window limit belongs to one conversation. The fix is a new chat, and there is no waiting period. A usage or message limit belongs to a period of time: you have used your allowance of messages, image generations, or file uploads, and ChatGPT will show a reset time. Waiting, or switching to another available model, is the fix there. A tool limit — for uploads, voice, image generation, or data analysis — is separate from text chat, and on Free and Go plans a tool limit does not stop everyday text chats.

OpenAI says that reaching an allowance does not by itself mean your account has been restricted, and a usage limit does not end your subscription. If you are unsure which limit you hit, read the notice itself: a "Context window full" or "Chat memory full" message points at the conversation, while a message with a reset time points at an allowance.

Frequently Asked Questions

Does deleting other chats free up space?

No. The limit is on the current conversation, not on your account. OpenAI's help pages say you do not need to clear other chats, and you do not need to turn off Memory or delete saved memories to fix a full conversation.

Will I lose my conversation when it fills up?

No. The conversation stays in your history exactly as it was. You can read it, copy from it, and search it. You just cannot send more messages in that thread.

Does upgrading stop this from happening?

It delays it. Paid plans have larger windows, so the same conversation lasts longer before filling up. But every plan has a limit, and a long enough conversation will fill any window. The summary-and-restart workflow is the permanent answer; a bigger plan is a useful cushion.

Why did a short chat hit the limit?

This has been reported by some users, and OpenAI Support has said it cannot always confirm a defect from those reports alone. Possible causes include a very long pasted message, large file uploads, or heavy tool output in a small number of turns. Start a clean new chat with only what you need, and contact OpenAI Support if it keeps happening.

Can I check how full my window is?

ChatGPT itself shows no token counter and no warning as a conversation fills, on any plan. Some third-party browser extensions add a gauge that estimates the fill level, but they are not part of ChatGPT and their estimates are approximate. The practical substitute is watching for the warning signs — repetition and forgotten context — and summarising early.

The Bottom Line

A full context window is not a punishment and not a malfunction. It is the predictable result of using a tool with a fixed working memory as though it had an infinite one. The conversation that fills up has done its job; your work now is to extract what matters from it, carry a short summary into a fresh chat, and keep going.

Start a new chat per task, ask for summaries before you need them, paste less and reference more, and keep a handover note of your own. Do that, and the "Context window full" message becomes a routine checkpoint instead of a lost afternoon.

Sources: OpenAI Help Center, "Why does ChatGPT say 'Context window full' or 'Chat memory full'?", as summarised in SSD Nodes' guide to the notice; SSD Nodes on ChatGPT limits and what to do when one is reached; OpenAI pricing page figures as read on 7 November 2026, via SSD Nodes; Chat AI Guide on ChatGPT context window sizes by model.