ChatGPT Context Window and Token Limits by Plan (2026)

ChatGPT's context window depends on your plan and on whether the answer comes from an instant or a reasoning model. As of September 2026, OpenAI's plan table gives instant models a 27K token window on Free, 54K on Go, Plus and Business, and 128K on Pro and Enterprise. Reasoning models get 256K on Go, Plus, Business and Enterprise and 400K on Pro. Those are total windows, input plus output. The API is a different world: GPT-6 Astra and the GPT-5.6 family accept 1,050,000 tokens with a 128,000 token output cap.

This guide lists the ChatGPT token limit for every plan, the API context window and max output for every current model, how OpenAI counts a token, what happens when a conversation hits the wall, and how to see how full your window is before it fills. Every figure below is read from OpenAI's own plan comparison, help center and model pages, and the date each was checked is stated. Sister guides cover the Claude context window and the Gemini context window.

What a Context Window Is in ChatGPT

A context window is the maximum number of tokens ChatGPT can hold at once for a single conversation, counting your messages, any files you attach and the model's own replies. When the total passes the window, the oldest tokens drop out and the model no longer sees them, which is why a long chat starts contradicting things you said an hour ago. OpenAI's plan table labels the figure "total context window" for exactly that reason: it is not an input limit alone.

A token is a chunk of text, usually a word or part of a word. OpenAI's help center gives the rule of thumb for English: one token is about four characters or three quarters of a word, so 100 tokens is roughly 75 words (OpenAI Help Center). Code, other languages and heavy punctuation tokenize less efficiently. For an exact count before you paste something large, use the free token counter.

Two things a context window does not do. It does not give ChatGPT permanent memory: ChatGPT Memory is a separate system that carries facts between chats. And it does not cap how many messages you can send per day; those are usage limits, covered in the ChatGPT limits guide.

ChatGPT Context Window by Plan (September 2026)

OpenAI publishes two windows per plan, one for instant models and one for reasoning models, and the reasoning window is four to five times larger on paid plans. The table below is read from openai.com/chatgpt/pricing and openai.com/business/pricing on September 9, 2026. The "pages" figures are OpenAI's own input maximums, which run lower than the total window because the window also has to hold the reply.

Three details are easy to miss. Go and Plus share the same windows; the $12 difference buys higher usage caps and GPT-6 Astra, not more context. Pro is the only plan with the 400K reasoning window that gets quoted as "the ChatGPT context window" in most articles, and Pro starts at $100 for five times Plus usage or $200 for twenty times. And Free has no fixed reasoning window because Free users cannot pin a reasoning model, so a Free account effectively lives inside 27K.

Which Model You Get on Each Plan

The window you get also depends on which model answers, and each plan unlocks a different set. On September 9, 2026 OpenAI's plan table names GPT-5.6 Luna for Free; Luna and GPT-5 Thinking Mini for Go; GPT-6 Astra, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna and GPT-5 Thinking Mini for Plus; and the same set plus GPT-5.6 Sol Pro for Pro. Business and Enterprise list the GPT-5.6 family, Sol Pro and Thinking Mini.

In practice, an instant answer from Luna on Plus runs inside 54K, and a reasoning answer from Sol or Astra on the same plan runs inside 256K. Switching the model mid-conversation changes the ceiling. For the full lineup and what each model is for, see ChatGPT models explained and the guide to reasoning effort in ChatGPT.

API Context Window and Max Output by Model

Through the API, GPT-6 Astra, the GPT-5.6 family, GPT-5.5 and GPT-5.4 all accept 1,050,000 tokens and return up to 128,000 tokens in one response. The figures below are from OpenAI's developer model pages, checked September 9, 2026. Pricing is per one million tokens and applies to API usage, not to ChatGPT subscriptions.

Two caveats from the model pages. On every 1,050,000 token model, input above 272,000 tokens is billed at twice the input rate and 1.5 times the output rate, so a full-window prompt costs more than the headline price suggests. And the API window is not what you get in the ChatGPT app: the same GPT-5.6 Sol that takes 1.05M tokens through the API answers inside a 256K window on Plus. Estimate a real bill with the AI cost calculator or compare models side by side in the model comparison tool.

How Many Words Fit in Each Window

Using OpenAI's rule of 75 words per 100 tokens, 27K tokens is about 20,000 words, 54K is about 40,000, 128K is about 96,000, 256K is about 192,000 and 400K is about 300,000 words of combined conversation. A full novel is 80,000 to 100,000 words, so a Plus reasoning window can hold two of them, while a Free instant window holds a long report.

Remember that the reply counts too. If you paste 35,000 words into a 54K window and ask for a 5,000 word rewrite, the request itself is already near the ceiling before ChatGPT answers. OpenAI's own page estimates are more conservative for that reason: it quotes about 40 pages of input for a 54K window, not 80. When you are close, split the job or use the chunking approach in the ChatGPT character and word limit guide.

Long ChatGPT threads, no gauge? AI Toolbox (formerly ChatGPT Toolbox) adds a free context window meter to ChatGPT that reads your plan's window and shows how full the conversation is, plus folders, full-text search and TXT export so nothing scrolls out of reach. 40,000+ users, 4.7/5 on the Chrome Web Store. Install AI Toolbox free ->

What Happens When a Conversation Hits the Limit

ChatGPT handles a full window in two ways: silently, by dropping the earliest messages from what the model sees, or loudly, with the message "You've reached the maximum length for this conversation." The silent case is the common one. Nothing is deleted from your chat history, the model simply stops reading the top of the thread, and answers start to drift.

The loud case ends the thread. You can still read it, but you cannot send another message, and the only native path is a new chat with whatever context you paste by hand. The fix, the warning signs and how to carry the conversation forward are in how to fix "maximum length for this conversation". If a chat you need has gone missing rather than gone quiet, start with the history recovery guide.

How to See How Full Your ChatGPT Context Is

ChatGPT shows no token counter and no warning as a conversation fills, on any plan. AI Toolbox adds a small context window gauge to the chat view that counts the tokens in the open conversation, compares them with your plan's window for the model in use, and shows the percentage used: green below 70 percent, amber from 70, red from 95, with a hint to hand the thread off before it fills. The meter is free; the paid tier adds a one-click handoff that summarises the thread and continues it in a fresh chat.

1. Install AI Toolbox from the Chrome Web Store on Chrome, Edge, Brave, Opera or Arc, then open chatgpt.com.
2. Open any conversation. The gauge appears in the chat view and updates as you type and as ChatGPT replies.
3. When the meter turns amber, export the thread or start a new chat and pull the old one in with @@ so the context comes with you.
4. Upgrade at ai-toolbox.co/pricing if you want the one-click handoff, unlimited Context Mentions and bulk export.

Who Hits the Window First

Developers hit it first. A pasted repository or a long stack trace tokenizes badly: code carries symbols and whitespace that push the tokens-per-word ratio well past OpenAI's English estimate, so a 54K window fills with far fewer than 40,000 words. Paste the failing module, not the project.

Researchers and analysts hit it on attachments. A 60-page PDF is roughly 30,000 words, which is already most of a Plus instant window before the first question. Use a reasoning model for document work; on Plus that moves you to 256K.

Writers and consultants hit it over time. A drafting thread that runs for weeks quietly passes 54K and the model forgets the brief. Folders and full-text search in AI Toolbox let you keep one chat per chapter or per client and find the right one instead of scrolling one giant thread; the organization guide covers the setup.

Limitations and What Native ChatGPT Handles Well

For most single-document tasks the native window is enough, and on Pro or a reasoning model it is generous. Native ChatGPT reads a long PDF, summarises a meeting transcript and holds a day's worth of back and forth without any extension. Projects also keep files and instructions available across chats, which reduces how much you need to re-paste.

The limits are the ones this guide has described. The app never shows how full the window is. The windows OpenAI publishes are estimates that vary with language and formatting, so treat 75 words per 100 tokens as a guide, not a contract. The meter's own token count is a close heuristic rather than OpenAI's exact tokenizer, and it cannot see the hidden system prompt or tool output that also occupy the window. Nothing in this guide changes the window itself: no extension can make a Free account hold more than 27K tokens.

Frequently Asked Questions

What is ChatGPT's context window in 2026?

It depends on the plan and the model type. As of September 2026, instant models get a 27K token window on Free, 54K on Go, Plus and Business, and 128K on Pro and Enterprise. Reasoning models get 256K on Go, Plus, Business and Enterprise and 400K on Pro. The figures are OpenAI's total windows, so they include the reply. Through the API, GPT-6 Astra and the GPT-5.6 models accept 1,050,000 tokens.

What is the ChatGPT Plus token limit?

On ChatGPT Plus the total context window is 54K tokens for instant models and 256K tokens for reasoning models, per OpenAI's plan table as of September 2026. OpenAI estimates that at about 40 pages of input for instant answers and 320 pages for reasoning answers. Go has the same two windows; Plus adds GPT-6 Astra and higher usage caps rather than more context.

Does ChatGPT Free have a token limit?

Yes. Free accounts get a 27K token total context window, which OpenAI estimates at about 12 pages of input. Free users cannot pin a reasoning model, so the reasoning window is listed as "varies" and the 27K figure is the practical ceiling. That is roughly 20,000 words of conversation, including ChatGPT's own replies.

What is the GPT-6 Astra context window?

Through the API, GPT-6 Astra has a 1,050,000 token context window and a 128,000 token maximum output, and inside ChatGPT it answers within your plan's reasoning window: 256K on Plus, Business and Enterprise, 400K on Pro. OpenAI began rolling Astra out to Plus, Pro, Business and Enterprise on September 4, 2026; it is not on Free or Go.

How many words is 54K tokens?

About 40,000 words of English, using OpenAI's rule of thumb that 100 tokens is roughly 75 words. Code, other languages and heavy formatting use more tokens per word, so the real figure can be lower. OpenAI's own estimate for a 54K window is about 40 pages of input, which leaves room for the reply.

Why does ChatGPT forget earlier parts of a conversation?

Because the conversation has grown past the context window, so the earliest messages are no longer sent to the model. Nothing is deleted from your history; the model simply cannot see the top of the thread. Start a new chat and bring the essentials with you, or use AI Toolbox's @@ Context Mentions to reference the old chat, and its context window meter to see the wall coming.

Is the ChatGPT context window the same as the GPT-5 token limit?

No. GPT-5's API context window is 400,000 tokens with a 128,000 token max output, while the ChatGPT app caps every model at your plan's window. API figures describe what a model can accept when you call it directly. Inside ChatGPT the plan decides: 27K on Free, 54K on Go and Plus, 128K on Pro for instant answers, and up to 400K for reasoning answers on Pro.

Bottom Line

The ChatGPT context window in 2026 is a plan feature, not a model feature: 27K on Free, 54K on Go, Plus and Business, 128K on Pro and Enterprise for instant answers, and 256K to 400K for reasoning answers on paid plans, all counted as input plus output. The API's 1,050,000 token windows on GPT-6 Astra and GPT-5.6 do not carry into the app. Since ChatGPT never shows how full the window is, add the free context window meter from AI Toolbox, keep one chat per task, and export or hand off a thread before the wall rather than after it.

Related reading: ChatGPT usage limits, ChatGPT pricing, Context Mentions with @@ and how to export ChatGPT conversations.

References

• OpenAI, ChatGPT pricing and plan comparison (context windows per plan, read September 9, 2026)
• OpenAI, Business and Enterprise pricing (Business and Enterprise windows)
• OpenAI developer docs, model pages (API context windows, max output, pricing)
• OpenAI Help Center, what are tokens and how to count them

Last updated: September 9, 2026