Your model benchmark measures everything except the bill

Every model benchmark I read ranks the same two things: how good the output was, and sometimes how...