What does it actually mean when people say an AI model has a 'context window'?

A context window is the amount of text — measured in tokens, not words — that an AI can "see" and remember during a single conversation or task before it starts forgetting things. Imagine you're having a conversation with someone who has a very specific kind of short-term memory. They can remember the last 20 minutes of your chat perfectly, but anything before that simply doesn't exist for them. That's basically how a context window works. For example, if you paste a 50-page report into ChatGPT and ask a question about something on page 3, the AI can only answer if page 3 still fits inside its context window. If your conversation has gone on too long and pushed that page out, the AI will either make something up or tell you it can't find that information. Tokens are the key unit here — roughly, one token equals about three-quarters of a word. So a 128,000-token context window (like what some newer models offer) can hold around 96,000 words, or about the length of a 300-page novel. But here's the catch: just because the window is big doesn't mean the AI pays equal attention to everything in it. Research from Anthropic and others has shown that AI models tend to focus more heavily on the beginning and end of the context, with a mushy middle that gets less attention. So if you're working with a long document, put your most important instructions or reference material either right at the start or just before your final question. **Related**: How many pages of text fit in a 128K context window? | Why does AI forget things I told it earlier in the chat?