| OpenAI o3 | GPT-5 | |
|---|---|---|
| arenaElo | 1418 | 1412 |
| humanEval | 92.4 | 91.6 |
| sweBench | 69.1 | 65 |
| mmlu | 90.2 | 91.5 |
| gpqa | 83.3 | 63.8 |
| aime | 91.6 | 92 |
| Price in/out ($/M) | $10 / $40 | $10 / $40 |
o3 reasons deeper but thinks for seconds per reply; GPT-5 is multimodal, faster and slightly better on MMLU. Route hard reasoning tasks to o3, everyday chat and multimodal to GPT-5.