But I don’t want to tokenmax. Do you have a solution that conserves tokens as much as possible, while still getting good results?