Context window
How much a model can hold in mind at once — everything you send plus everything it writes back, counted in tokens.
how it works · the vocabulary
windowlong context
Exceed it and the oldest part of the conversation falls out or the request is refused. A million-token window sounds like it removes the problem, and it does not: you pay for every token you put in it on every single call, so sending a whole codebase each turn is a bill rather than a technique. The useful question is not how large the window is but how much of it you must refill each time.