Harness · anthropic (proprietary)
Claude Code
The strongest proprietary coding harness in the ring, and the honest ceiling to measure against. Included because a benchmark that only ranks the harnesses we can beat is not a benchmark.
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 →
Result: not yet measured
The facts
- Maintainer
- Anthropic
- Source
- anthropic (proprietary)
- License
- Proprietary
- Category
- Coding harness
- Headless run
- Yes — `claude -p` (print mode, `--output-format json`)
- Native OpenRouter
- — can be pinned to one shared model
- Determinism notes
- 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.
Where it sits
Performance scores land here only with published traces, under the pre-registered method. Until then, this profile is the fair-run readiness check: can we start Claude Code headless and pin it to one shared model to isolate the scaffold?