Head to head
Codex vs Claude Code
The facts side by side — model held constant. No performance verdict until traces are published under the pre-registered method.
AEQI builds one of the harnesses measured here. This board is in method preview and pre-registration — no performance scores are published yet, and we lead with our own losses, not our wins. Read the method & charter →
| Attribute | Codex | Claude Code |
|---|---|---|
| Maintainer | OpenAI | Anthropic |
| License | Apache-2.0 | Proprietary |
| Category | Coding harness | Coding harness |
| Headless run | Yes — `codex exec` (non-interactive, `--json` event stream) | Yes — `claude -p` (print mode, `--output-format json`) |
| Native OpenRouter | ||
| Determinism | Pin the npm version. Points at any OpenAI-compatible base URL via `model_providers`, so one pinned model is straightforward. | Speaks the Anthropic wire format rather than the OpenAI one, but OpenRouter serves that format too — so it runs on the same pinned model and the same account as every other entrant, with no proxy in between. |
| Benchmark result | Not yet measured | Not yet measured |
Questions
- Is Codex better than Claude Code?
- The AEQI Harness Index does not publish a performance verdict for Codex vs Claude Code yet — scores land only with published traces under the pre-registered method. What this page compares are the verifiable facts: license, category, headless capability, and whether each can be pinned to one shared model for a fair run.
- Can Codex and Claude Code be benchmarked fairly against each other?
- Yes — both document native OpenRouter support, so we can point them at one identical pinned model and attribute any difference to the scaffold, not the model.