The difference in their post training means that base models and reasoning models produce different outcomes.