Gotta assume theyre reducing overall compute with smaller models cause 200$ aint squat for their investment.