Codex won a real-work benchmark after one config-file change
A practical coding benchmark used real Python and SQLite tooling, older code that needed fixes, planted bugs, and five staged change requests. , , Codex, and Gemini each received the same prompt in their own CLI and had one fully automatic attempt. Codex with default settings scored 91.1 and worked quickly, but it stopped short on a hard step that the s solved better.
Telling Codex to keep optimizing made things worse because it broke the and scored 88.9. A S markdown file changed the result by spelling out habits such as small focused edits, finding the real cause, adding a , treating every stated limit as firm, and checking the work before calling it done. With that file, Codex scored 97.1, finished fastest, and cost least; a repeat run scored 96.1.
The practical lesson is that real older-code tasks may reveal more than clean new-project tests, and a strong instruction file can matter more than paying for a higher model tier. still showed a reason for its higher price because it found provably best answers inside the limits without extra prompting on the hardest algorithm work.
Key points
- The benchmark compared , , Codex, and Gemini on real Python/SQLite work and older-code fixes.
- Default Codex scored 91.1 and was fast, but it missed the best answer on a hard task.
- A simple “try harder” prompt broke the and lowered the Codex score to 88.9.
- A S markdown file lifted Codex to 97.1, with a repeat score of 96.1.
- still stood out for hard algorithm work where the best answer must be found inside strict limits.