| GPT-5.4 Codex | GPT-5 Codex | |
|---|---|---|
| arenaElo | 1408 | 1395 |
| humanEval | 95.5 | 94.8 |
| sweBench | 70 | 68 |
| mmlu | 88.6 | 87.2 |
| gpqa | 60.8 | 58.4 |
| aime | 86 | 85 |
| Price in/out ($/M) | $9 / $36 | $8 / $32 |
5.4 Codex gains ~0.7 HumanEval points, a 100K bigger context and ~2 extra GPQA points over 5 Codex, at ~12% more cost. For high-stakes automated PR pipelines, worth the bump. For routine autocomplete, 5 Codex still wins on price.