[skywhopper]

Yeah, the point is that this behavior uses a lot more tokens than the OP says is a “typical” LLM query.