[Davidzheng]

Am i misunderstanding? spending more tokens certainly is not optimizing for compute cost. It's the opposite