[golem14]

That does not help you compare pricing between models since they can vary by (even internal, thinking) output size.

The best way to measure is really the end-2-end cost, price per task.