OpenAI code refactoring tasks - reported
Code · 2025-09-15
OpenAI's internal evaluation of large refactor-style tasks drawn from established repositories in Python, Go and OCaml (for example a Gitea pull request touching 232 files), run agentically in Codex. Task set, grading and run count are unpublished; diagnostic. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5-Codex | 51.3 |
| GPT-5 | 33.9 |