o3-mini-high runs the same reasoning engine as o3-mini but with its thinking effort dialed up, meaning it spends more compute per problem before responding. This makes it noticeably stronger on multi-step math, logic puzzles, and code that requires careful planning — though that extra deliberation comes with slower response times. It behaves like a methodical colleague who insists on double-checking their work before speaking.