| OpenAI o4-mini | Claude Haiku 4.5 | |
|---|---|---|
| arenaElo | 1362 | 1274 |
| humanEval | 88.6 | 78.2 |
| sweBench | 60 | 40.6 |
| mmlu | 85.4 | 81.5 |
| gpqa | 74.2 | 48 |
| aime | 92 | 60 |
| Price in/out ($/M) | $1.1 / $4.4 | $1 / $5 |
o4-mini reasons dramatically better (GPQA 74 vs 48) but is slower and pricier per request than Haiku. Haiku is the default for classification and short chat; o4-mini is the smart pick when problems need real thinking.
OpenAI o4-mini · Claude Haiku 4.5 · All AI model comparisons