Codex performed better in a small embedded refactor test
A small embedded project was from AsyncTCP and AsyncWebserver to ESP-IDF’s built-in esp_http_server. The job was to serve a few web pages, including an index page, firmware upload and flashing, and a Wi-Fi portal. The plan, checklist, and prompt map were prepared with Codex 5.5 xhigh.
Free was tried first through Opencode because the task looked simple and the goal was to save for another project. After flashing the Deepseek result to the device, it caused a boot loop. The issue was fixed and the project became functional, but the quality was still not satisfying.
The same prompt was then run with Codex 5.5 xhigh through Codex CLI, and that was judged much better. Opus 4.8 was also tested through Claude Code using tokens, with Codex used afterward to compare and rate the three s.
Key points
- The task moved a small embedded project from AsyncTCP and AsyncWebserver to ESP-IDF’s esp_http_server.
- Free produced an initial result that caused a boot loop after flashing.
- The Deepseek version became functional after fixes, but the quality was considered weak.
- Codex 5.5 xhigh produced a much better result from the same prompt.
- Opus 4.8 was also tested through Claude Code and included in the comparison.