will be a combination of various factors. How much reasoning models you use + the weaker models and the type of tasks you perform and finally the token cost at that time.

kind of hard to predict