Comment by Turskarama

They don't want to waste tokens on purpose, what they're actually hiding is when the model wastes tokens on obviously stupid "thoughts".