[chatmasta]

Token costs are not zero when you’re running local models, because you paid for the hardware, and you can’t scale inference indefinitely without paying for more hardware.