Context Window

A context window is the fixed amount of text a model can read on a single call, measured in tokens (a token is roughly three quarters of a word). Everything the model should use, the question, the instructions, any documents, the conversation so far, has to fit inside it. Anything outside the window does not exist to the model on that call.

Two consequences follow. First, a model has no memory between calls: it sees only what is in the window this time, so anything from last turn has to be put back in. Second, a bigger window is not more memory: models read the start and end of a long prompt more reliably than the middle, so where a fact is placed matters as much as whether it is present.