What is a context window?

The context window is an AI model's working memory for a single conversation or task. Everything the model considers when producing a response has to fit inside it: the instructions, earlier messages, pasted text, uploaded documents, and the reply being written. Its size is measured in tokens, the small chunks of text that models read, and different models have different limits.

This matters in practical ways. A long contract plus a lengthy back-and-forth about it might fit comfortably in one model's context window and not in another's. For example, someone asking an assistant to compare several long reports may find that the tool can only use part of the material, or rejects the upload, because together the documents exceed what the model can hold.

When a conversation grows past the limit, older material has to be dropped or condensed, which is why a long chat can seem to forget instructions given at the start. Even within the limit, models can pay less attention to details buried in the middle of a very long input, so a larger window does not guarantee that every detail will be used.

Practical habits help: start a new conversation for a new task, restate key instructions when a chat gets long, and share only the relevant sections of large documents. Tools that search a document collection and pass in just the matching passages, an approach known as retrieval-augmented generation, are another way to work with more material than fits in the window.

An example

A long chat about a project starts ignoring the formatting rules set in the first message, because that message has fallen outside the context window.