| OpenAI o4-mini | DeepSeek R1 | |
|---|---|---|
| arenaElo | 1362 | 1389 |
| humanEval | 88.6 | 90.2 |
| sweBench | 60 | 49.2 |
| mmlu | 85.4 | 90.8 |
| gpqa | 74.2 | 71.5 |
| aime | 92 | 79.8 |
| Price in/out ($/M) | $1.1 / $4.4 | $0.55 / $2.19 |
DeepSeek R1 still wins on GPQA and is open-weights; o4-mini is better integrated, slightly faster and easier to productionize. R1 for research and self-host; o4-mini for polished cloud deployments.