What Is a Context Window? Why AI Forgets and How to Work Within the Limit
What is a context window? It is the amount of text an AI model can consider at one time: your instructions, the conversation so far, any documents you uploaded and the reply it is writing. Everything outside that window is invisible to the model. Understanding the context window explains why long chats drift, why huge documents get summarized badly and why starting a fresh conversation often fixes strange answers. This explainer covers tokens, real limits on popular plans, symptoms of hitting the limit and practical ways to work within it.
Tokens: the unit of context
AI models do not count words; they count tokens. A token is a chunk of text, often a word or part of a word. Both what you send and what the model writes count toward the context window.
Rules of thumb:
- Short common words are usually one token.
- Long or unusual words split into several tokens.
- Code, tables and non-English text can use more tokens than you expect.
What is a context window made of?
At any moment, the window holds:
- System instructions set by the product.
- Your custom instructions or project instructions.
- The conversation history that still fits.
- Uploaded or retrieved documents.
- The reply being generated.
When the total exceeds the limit, something has to give: older messages may be dropped or summarized, or the request may fail.
How big are context windows today?
Limits vary by model and plan. As one example, OpenAI's ChatGPT pricing page listed the following context windows for its Instant model in October 2026:
| Plan | Context window (GPT Instant) |
|---|---|
| Free | 27K tokens |
| Go and Plus | 54K tokens |
| Pro | 128K tokens |
Some API models advertise much larger windows; DeepSeek's API documentation, for example, lists a 1M-token context for its models. Bigger windows help with long documents, but they do not guarantee the model pays equal attention to everything inside them.
Symptoms you are hitting the limit
- The AI forgets instructions you gave at the start.
- Answers contradict earlier decisions.
- Summaries skip whole sections of long documents.
- Errors or truncated replies when uploading large files.
- Quality drops in very long conversations.
Working within the context window
Start fresh with a summary
When a chat gets long, ask: "Summarize the key decisions, facts and open questions from this conversation in 200 words." Start a new chat and paste the summary.
Use projects for stable context
Projects keep files and instructions available without pasting them every time. See our ChatGPT Projects guide and Claude Projects tutorial.
Chunk long documents
Split long documents into sections, summarize each, then ask questions across the summaries.
Be selective
Upload only the parts of a document you need, and remove boilerplate, appendices and repeated tables.
Put key instructions near the request
Restate important constraints in the message where they matter, rather than relying on instructions from much earlier.
Context window vs memory vs retrieval
- Context window: what the model can consider right now.
- Memory: details saved across conversations, injected into context when relevant.
- Retrieval: searching documents and adding only relevant parts to the context.
Retrieval is how assistants handle large knowledge bases without huge windows.
Context windows and cost
On APIs, you typically pay per token for input and output. Sending large contexts on every request costs more. Trim inputs and reuse summaries to save money. For budgeting AI usage, see our guide to what AI automation costs, and for how plans differ between assistants, our Grok vs ChatGPT comparison.
Practical examples
Long contract review: a 60-page contract may exceed the context window on smaller plans. Split it into sections, ask for key terms per section, then ask for an overall summary from those notes.
Ongoing project chat: after weeks of messages, the assistant forgets early decisions. Ask it to summarize the project status, start a fresh chat in the same project, and paste the summary.
Big spreadsheet: instead of pasting thousands of rows into chat, use a data analysis feature that reads the file directly, or summarize the data first.
Customer chatbot: rather than stuffing every help article into the prompt, use a tool that retrieves only the relevant articles for each question.
Choosing a plan with context in mind
If you regularly work with long documents, compare context limits as well as price when choosing an AI plan. Larger windows on higher tiers can save time, but good habits such as chunking and summarizing often matter more than raw window size.
Common misconceptions
- "Bigger windows mean perfect recall." Models may still pay less attention to the middle of very long inputs.
- "The AI remembers everything I've ever told it." Only what fits in the window, plus any separate memory features.
- "Uploading a file means the AI read all of it." Some tools retrieve only parts of large files.
Context windows in AI agents
AI agents that run long tasks, such as research or multi-step coding, juggle large amounts of context: instructions, tool results and intermediate notes. Agents often summarize their own progress to stay within limits. If an agent loses track of instructions partway through a long task, break the task into smaller pieces or restate the key requirements at each stage.
A quick habit
At the end of any long working session, ask the AI for a short summary of decisions and next steps, and save it in your notes. You will have a clean starting point for the next session, even if the old conversation has drifted or grown too long to be useful.
Related reading
For more tools grouped by job, browse our AI tools by profession guide.
FAQ
What is a context window in simple terms?
It is the maximum amount of text, measured in tokens, that an AI model can consider at once, including your messages, any files and its own replies.
What is a token?
A token is a chunk of text, often part of a word. Models count both input and output in tokens.
How big are context windows?
It varies by model and plan. For example, OpenAI's ChatGPT pricing page listed context windows for its Instant model of 27K tokens on Free, 54K on Go and Plus, and 128K on Pro in October 2026. Some API models advertise windows of a million tokens.
Why does the AI forget what I said earlier?
In long conversations, earlier content can fall outside the context window or get less attention. Starting a new chat with a summary often helps.
Is the context window the same as memory?
No. The context window is what the model considers in one conversation. Memory features save details across conversations separately.
Conclusion
What is a context window? The limit on how much an AI can consider at once, measured in tokens. Once you understand it, odd behavior in long chats makes sense. Summarize and restart long conversations, use projects for stable context, chunk big documents and keep key instructions close to your request.
Sources and methodology
ChatGPT context window figures come from OpenAI's pricing page (https://openai.com/chatgpt/pricing/), and DeepSeek's 1M-token context from its API pricing documentation (https://api-docs.deepseek.com/quick_start/pricing), both verified 14 October 2026. Limits change frequently.