| Claude Code | GPT-5 Codex | |
|---|---|---|
| arenaElo | 1420 | 1395 |
| humanEval | 93.1 | 94.8 |
| sweBench | 64.3 | 68 |
| mmlu | 92.3 | 87.2 |
| gpqa | 65.2 | 58.4 |
| aime | 88 | 85 |
| Price in/out ($/M) | $15 / $75 | $8 / $32 |
Claude Code is the CLI, GPT-5 Codex is the model. If you want OpenAI's coder inside an Anthropic-style CLI, combine Codex with Codex CLI. Compared directly: Claude Code has the better UX and agent stability, GPT-5 Codex has the better raw HumanEval.