🤷🏻‍♂️ i haven't used codex. in general i find that openai benchmarks look better than the model performs