Every model benchmark I read ranks the same two things: how good the output was, and sometimes how many tokens it took. Almost none of them print the number you actually get charged for.
That gap matters, because the two rankings are not the same list. A model can win on quality, lose on tokens, and still be the cheapest way to finish your task — or the opposite. You cannot tell which without doing the multiplication yourself.
Here is the multiplication.
The formula
cost = (input_tokens / 1M) × price_in + (output_tokens / 1M) × price_out
