o4 mini high

o4-mini-high is designed for tougher prompts where correctness matters more than raw throughput. It plans steps more carefully, checks intermediate results, and adheres closely to requested formats, which improves performance on multi-step math, complex refactors, and policy-aware drafting. You get the same integration surface as o4-mini, including retrieval grounding and function calls, but with a more deliberate decoding profile. Teams route routine traffic to o4-mini and reserve o4-mini-high for queries that need extra reasoning headroom.