| Claude Sonnet 4.6 | GPT-5 mini | |
|---|---|---|
| arenaElo | 1358 | 1288 |
| humanEval | 89 | 81 |
| sweBench | 60 | 50 |
| mmlu | 88.7 | 82.4 |
| gpqa | 59.4 | 50.1 |
| aime | 80 | 78 |
| Price in/out ($/M) | $3 / $15 | $0.25 / $2 |
Sonnet 4.6 is the stronger model on reasoning and coding; GPT-5 mini is dramatically cheaper ($0.25/M vs $3/M input). Use GPT-5 mini for high-volume UX, Sonnet for anything user-facing where quality matters.