Context window
Also called: context length, context
In one sentence
The maximum amount of text, measured in tokens, a model can consider at once.
In more depth
The window holds everything the model can see: system instructions, the conversation so far, any documents supplied, and the response being generated.
When a conversation exceeds the window, something must be dropped or summarised — which is why long conversations can appear to forget earlier details. A larger window costs more per request and does not guarantee the model attends equally to everything inside it.