Context window.

how much text a model can consider at once (prompt plus its reply), measured in tokens. Bigger windows let it work over longer documents and conversations.