Code Edit Format Benchmark
February 27, 2026 ยท View on GitHub
Benchmark guide for comparing edit format performance and render behavior.
Purpose
Use this benchmark to compare edit-format throughput and output quality for Codex-focused editing workloads.
Quick Start
npm run bench:edit-formats
Smoke run:
npm run bench:edit-formats:smoke
Render dashboard output:
npm run bench:edit-formats:render
Output Files
| Output | Location |
|---|---|
| Benchmark report JSON | .tmp/edit-format-benchmark-*.json |
| Render preview artifacts | .tmp/edit-format-benchmark-render-*.txt |
(Temporary benchmark artifacts are not source files.)
Common Presets
| Preset | Goal |
|---|---|
codex-core | Baseline Codex-oriented evaluation |
smoke | Fast sanity check for CI/local validation |
Interpretation Checklist
- Compare latency per format.
- Measure token/size overhead.
- Review success/error rates.
- Validate output consistency.
- Confirm no regressions in editing fidelity.
Cleanup
Bash:
rm -rf .tmp
PowerShell:
Remove-Item ".tmp" -Recurse -Force -ErrorAction SilentlyContinue