What Is a Context Window? Tokens Explained Simply

Ever had a long chat with an AI where it suddenly forgot something you said at the start? Or pasted in a big document and got an error saying it’s too long? Both come down to one idea: the context window.

It’s one of the most useful AI concepts to understand, because it explains a lot of strange behaviour and tells you how to work around it.

The short version

A context window is the amount of text an AI model can look at in one go. That includes everything:

• your current message
• earlier messages in the conversation
• the AI’s own previous replies
• any documents, files or web pages it has read
• hidden instructions from the app you’re using

Think of it as the model’s desk. Everything it needs has to fit on the desk at once. Whatever doesn’t fit, it can’t see.

What’s a token?

Context windows are measured in tokens, not words. A token is a chunk of text, often a short word or part of a longer one. “Cat” might be one token; “unbelievable” might be split into several.

A handy rule of thumb for English:

• 1 token ≈ 0.75 words
• 1,000 tokens ≈ 750 words, roughly a page and a half of text
• 100,000 tokens ≈ 75,000 words, about the length of a novel
• 1,000,000 tokens ≈ 750,000 words, several long books

These are rough figures. Code, numbers, unusual words and many non-English languages tend to use more tokens per word, so the same window holds less of them.

How big are context windows now?

They’ve grown fast. A few years ago, a few thousand tokens was normal. Today, many popular models handle hundreds of thousands of tokens, and several newer models accept around a million. In October 2026 alone, several new models launched with million-token windows at low prices. See our AI This Week roundup for examples.

The app you use may allow less than the underlying model supports, especially on free plans. Check your plan’s details if you work with big files.

Why chats “forget” things

In a long conversation, every message adds to the pile. When it no longer fits, the app has to do something. Depending on the tool, it may:

• drop the oldest messages
• replace earlier parts with a shorter summary
• refuse new input until you start a new chat

Either way, the instruction you gave 80 messages ago may no longer be on the desk. The model isn’t ignoring you; it can’t see it any more.

The fix: when a chat gets long, ask for a summary and start fresh.

Summarize this conversation so far: my goal, the decisions we've made,
my preferences and instructions, and what's left to do. Keep it under
300 words so I can paste it into a new chat.

Paste that summary into a new conversation and carry on.

Bigger isn’t automatically better

A huge context window is useful, but it has trade-offs:

• Details can get missed. Models sometimes pay less attention to material in the middle of very long inputs. Newer models are better at this, but not perfect.
• Cost and speed. On APIs, you pay per token, and longer inputs take longer to process. Pasting a 300-page manual to ask one question is slow and expensive.
• Noise. Irrelevant material can distract the model from what matters.

For the best results, give the model what it needs and not much more. If you only need chapter 4, paste chapter 4.

Practical ways to work within the limit

1. Trim before you paste. Remove headers, footers, boilerplate and sections that don’t relate to your question.
2. Work in sections. For very long documents, summarize part by part, then combine. Our guide to summarizing long documents with AI shows this step by step.
3. Restate key instructions. In long chats, repeat important rules near the end: “Remember: British spelling, under 200 words.”
4. Use files or projects features. Many apps let you attach documents to a project so you don’t have to paste them every time.
5. Start new chats for new tasks. One topic per conversation keeps the window clean.

How to estimate tokens

To skip the maths, paste your text into our free Token Counter. for your own text

You don’t need exact numbers most of the time, but a quick estimate helps you decide whether to paste a whole document or just part of it.

1. Count the words. Most word processors show this at the bottom of the screen or in a “word count” menu.
2. Divide by 0.75. A 6,000-word report is roughly 8,000 tokens.
3. Add room for the reply. The model’s answer also uses the window, so leave space for it. A long report or a big block of code in the answer can use thousands of tokens.

Some AI providers offer free token counters or show token usage in their developer tools if you need precise figures.

Context window vs memory

These get confused. The context window is short-term: what the model can see right now in this conversation. Some apps also offer memory, which saves facts about you between conversations and adds them back into the context when relevant. Memory still has to fit in the window; it’s just chosen for you automatically.

Key takeaways

• The context window is how much text an AI can see at once, measured in tokens.
• Roughly, 1,000 tokens equals 750 English words.
• When a chat exceeds the window, older content gets dropped or compressed, so the AI “forgets”.
• Bigger windows help, but focused input still gives better, cheaper and faster answers.

Next, learn how agents use the context window to work through long tasks in what is an AI agent?.