One dev compares Codex, Gemini CLI, and Claude Code for everyday coding help
A compared several while looking for a practical alternative to , which they found underwhelming. Their main use cases were spotting bugs, tidying up functions, and generating starter code. has improved noticeably since they last tried it, enough to consider resubscribing.
Gemini CLI made more mistakes but still performed reasonably well overall. Claude Code could not be fully tested because the transition happened on the same day. Grok Code is still in beta, and Mistral has its own coding tool entering the picture.
The developer is also exploring locally-run models via tools like OpenCode to avoid being locked into any single provider, though results have been mixed so far. A past experience of Grok moralizing in a way reminiscent of ChatGPT 5.2 put them off that option entirely.
Key points
- dissatisfaction prompted a hands-on comparison of Codex, Gemini CLI, and Claude Code
- Codex has improved enough that the developer is considering paying for it again
- Gemini CLI stumbled but was decent overall
- Claude was skipped this round due to the transition happening the same day
- Locally-run models via OpenCode are being explored to avoid , with mixed results so far