| OpenAI o4 | OpenAI o3 | |
|---|---|---|
| arenaElo | 1438 | 1418 |
| humanEval | 94 | 92.4 |
| sweBench | 72 | 69.1 |
| mmlu | 92.5 | 90.2 |
| gpqa | 87.1 | 83.3 |
| aime | 94 | 91.6 |
| Price in/out ($/M) | $12 / $48 | $10 / $40 |
o4 tops o3 on every reasoning benchmark (GPQA 87 vs 83, MMLU 92.5 vs 90.2) at a ~20% lower list price. Unless you have o3 pinned by contract, o4 is a straight upgrade for reasoning-heavy agents.