I was doing some experiments to understand the cost structure of the assistant api and found out, that a lot of the additional, unexpected cost came from the context tokens. So I figured: Let’s delete all threads and start fresh tommorow so I can ask a single question (I used the playground) in order to see how much I will be billed.
I was billed around 0.12$ with 4 api requests listed in the api request tab. Why does it say 4, when I have only asked 1 question. Furthermore: Am I correct in my determination, that the context tokens are not just made of previous dialogue but also the document text the model considered for the answer? Because the context tokens were still quite high despite no previous dialogue…