Context window

The context window is the text the model is working with at this moment: your prompt, anything you pasted in, and the conversation so far. It is measured in tokens, roughly three quarters of a word each, so a 200,000-token window is about 150,000 words.

This explains a class of frustration that feels like stupidity. You ask for "our tone" and get generic copy because your tone was never in the window. You ask about a document you did not paste. You continue yesterday’s chat, which the model does not remember. There was nothing to remember. In a very long chat, material in the middle is also used less reliably than material at the start or the end.