Yes, Deepseek consumes a higher number of tokens per turn that comparable models (the latest Muse for example). This is just my experience, not a scientific study.