Does Vapi feed in the prompt every turn? I calculated my output and input tokens for the llm model I have and it should really only be like 4 cents per 2 minute call, but vapi is quoting almost 9 cents per MINUTE, is it marked up?