o3

o3 thinks before it speaks — literally. It runs extended internal reasoning chains before producing a response, which makes it noticeably slower but significantly more reliable on problems requiring multi-step logic, mathematics, or careful deduction. It handles ambiguous or hard problems by working through them rather than pattern-matching to a quick answer, though that deliberation comes at a higher compute cost.