Context window
Also called context length · context limit
The maximum amount of text, measured in tokens, that a model can hold in view for a single request.
In more detail
Everything the model can reason about on one call (your instructions, the conversation so far, any documents you pasted, and the response it is generating) has to fit inside the context window together. Exceed it and something has to be dropped or summarised. The window is a ceiling on what the model can see at once, not a memory that persists between calls.
Example
Asking a model to write an hour-long script in one request pushes against the window from both ends: the instructions and prior chapters take up input space, and the script itself takes up output space. Splitting the job into chapter-sized calls, each given a summary of what came before, keeps every individual request comfortably inside it.
Why it matters
This is the technical reason long-form scripts drift. A model that has run out of room for chapter two by the time it writes chapter nine will contradict itself, and no amount of prompt wording fixes it. Pipelines that produce coherent long-form work do it by managing what goes into each window, which is why they make several calls instead of one.