Costs in actual use

Showing you work is for suckers, but to me it looks like if your application needs an average of one word per second, then you will be paying about ten dollars a day using the most expensive model.

And it is worth using the cheapest model if you can get similar quality results with 75x more words per prompt, which is the difference between a single sentence prompt and maxing out the token limit with many examples.

Just thought that might help give people some perspective.