A series of individual tokens (words or subwords) that the model generates one after another to form a complete response.