Reasoning models generate a long internal chain of thought — exploring approaches, checking work, backtracking — and then give a final answer. They excel at math, coding and science problems but are slower and more expensive per question.
Many products let users choose how much “thinking effort” a model spends.