That's why you have to reframe in terms of total cost per task and factor in model token generation quantity and multiply that by the base cost of the model. Then factor in your time value if you dare. Then you should get a more meaningful business metric.