o4-mini-high runs o4-mini with its reasoning effort dialed to the maximum setting, meaning it spends more compute thinking through problems before responding. The result is noticeably stronger performance on math, logic, and code tasks compared to the default effort level, at the cost of slower responses. It behaves like a careful analyst who double-checks work rather than answering off the cuff.