Yes but with prompt caching decreasing the cost of the input by 90% and with output tokens not being cached and costing more than what do you think that results in?