look at token use, 3.8 flash is a huge token hog compared to openai models