| GPT-5.5 | GPT-5 | |
|---|---|---|
| arenaElo | 1432 | 1412 |
| humanEval | 94.2 | 91.6 |
| sweBench | 66 | 65 |
| mmlu | 93 | 91.5 |
| gpqa | 68.7 | 63.8 |
| aime | 92 | 92 |
| Price in/out ($/M) | $12 / $48 | $10 / $40 |
GPT-5.5 pushes Arena Elo past 1430, gains 50% more context (600K vs 400K), and sharpens every benchmark. Pricing rises ~20%. If you're on GPT-5 today, test 5.5 for high-value workloads first; swap en masse once your evals hold.