| Cursor Agent | GPT-5 Codex | |
|---|---|---|
| arenaElo | 1378 | 1395 |
| humanEval | 91.2 | 94.8 |
| sweBench | 62 | 68 |
| mmlu | 88.7 | 87.2 |
| gpqa | 59.4 | 58.4 |
| aime | 85 | 85 |
| Price in/out ($/M) | $0 / $0 | $8 / $32 |
Cursor routes to multiple models and shines on SaaS UIs; GPT-5 Codex is the raw coding brain you can call from your own pipeline. Startups shipping their first B2B SaaS should default to Cursor. Agencies automating PR generation across client SaaS repos should wire GPT-5 Codex directly.