Yeah. Some of it may just be "thinking" less rather than faster token generation.