Claude, Cody & Gemini
- published
To date, my only sizable experience with AI has been OpenAI’s ChatGPT and Codex. However, I was curious about the big alternatives, Gemini and Claude. I’ve heard on podcasts that Gemini is best for front-end work and Codex for back-end. So I gave them a shot. I’ve spent minimal time with each on the web interface and the CLI.
Here’s what I’ve tested so far:
- Codex: GPT-5.4 w/ extra high reasoning
- ChatGPT: GPT-5.4 w/ Thinking
- Gemini: Gemini 3 w/ Thinking
- Gemini CLI: Gemini 3 - v0.33.1
- Claude Code: Sonnet 4.6 - v2.1.76 - medium effort
Gemini has been very disappointing, but perhaps this is a me problem. I’ve grown accustomed to Codex prompting me about areas where I’ve been unclear. Gemini, on the other hand, will just roll with anything it is handed, even if the results suffer dramatically from the ambiguity.
Claude Code has been an interesting mix. It doesn’t always understand my ambiguous commands as accurately as Codex or ChatGPT. However, it comes up with solutions that are interesting forks of my instructions.
I may post here now and then about the experience.