| GPT-5.5 | Claude Opus 4.7 | |
|---|---|---|
| arenaElo | 1432 | 1420 |
| humanEval | 94.2 | 93.1 |
| sweBench | 66 | 64.3 |
| mmlu | 93 | 92.3 |
| gpqa | 68.7 | 65.2 |
| aime | 92 | 88 |
| Price in/out ($/M) | $12 / $48 | $15 / $75 |
GPT-5.5 edges Opus 4.7 on Arena Elo and HumanEval by a hair, while Opus 4.7 still holds the 1M context crown and long-horizon agent stability. For voice + multimodal, GPT-5.5. For coding agents and long-doc legal work, Opus 4.7.