I’ve been trying out the “auto” agent select this week. Overall quality results have been good, but there’s no indication of which model is being used. I decided to check my Cursor account logs, was surprised to see a much higher volume of requests than normal. So I scrolled through the Usage-Based Pricing log, and discovered that it has been using claude-3.7-sonnet-max
a LOT. This is frustrating for several reasons:
- Max is supposed to be for ■■■■■■■■ users who need to push the limits of the AI. I keep my projects and requests small and simple, the AI helps me with stuff and I don’t give it anything big to work on.
- Max is usage-based pricing ONLY, higher cost per request, AND it eats an additional request per tool call
- I never enabled the Max model in Cursor Settings and I’ve never used it intentionally
After perusing my logs, it seems whenever an “Agent” mode request is made, it’s almost always using the Max model. This feels pretty scummy to me that auto mode was just choosing to use the super expensive model and gobbling up my requests without any sort of indication to me. I’m glad I thought to look at my usage logs before it got too out of hand.