Context window (tokens)

The context window is the amount of text an AI model can work with in a single conversation or task. It usually covers your instructions, any documents you paste in, the conversation so far and the answer being written. It is measured in tokens, which are small pieces of text. In English, one token is roughly three-quarters of a word, so a window of a hundred thousand tokens holds a long report. In the library, this is recorded as a figure for each tool.

A small window forces you to split long documents into pieces, and the assistant can lose track of earlier parts of a conversation. A large window lets you ask questions about a whole contract, a long meeting transcript or a full set of notes in one go. You need a big window when you work with long files. For short questions and drafts, it makes little difference, and a larger window does not by itself guarantee accurate answers about everything in it.

• Ask what limit applies on the plan you would actually use, because it can differ from the figure on the model's page.
• Test the tool by pasting in your longest real document and asking about a detail from the middle.
• Check whether the limit includes the answer as well as your input.
• Find out what the tool does when you go over the limit, for example whether it cuts off the oldest text quietly.