That's max output tokens per response limit, separate from context length