I’m curious how people are using AI for coding in the terminal—and which tools are actually working well for them.
I’ve been trying OpenCode with free models so far, and I’ve had a really good experience: it’s responsive and has worked without any problems. Now I’d like to compare it with Codex and Claude Code.
If you use AI coding tools in the terminal, I’d love to hear about your experience:
- Which tool do you use: OpenCode, Codex, Claude Code, or something else?
- Which models have you tried, and are you using free or paid versions?
- What kinds of coding tasks do you use it for?
- How many back-and-forth interactions does it usually take to get a good result?
- What does your tool do especially well—or poorly?
- Any tips for someone deciding what to try next?
I’m especially interested in real-world comparisons: which tool gets you to a strong result with the least friction, and why?
Share your setup or a specific example if you can. I’d love to learn what’s working for you!
Top comments (4)
For terminal agents, the deciding factor is less the brand than the control surface: how tool permissions work, how context is recovered, whether changes are reviewable, and how easily a team can run the workflow again. A small task benchmark would make the comparison much more useful.
I’m reaching the same conclusion, though I’ve only tried OpenCode and GitHub Copilot. I used OpenAI and Claude models through Copilot and noticed some differences between them, but I imagine using each provider’s own terminal agent could feel different too.
So far, I’m enjoying OpenCode for both coding and planning. It’s responsive and works well for me.
The brand matters less than the harness. The history goes back to the model with every turn, so session length sets the price, not the logo. I built Tokenade, which sits between the agent and the model in Claude Code and OpenCode, and in most of Codex CLI.
Thats make sense, I'll continue using Opencode 😀
You and other person already said the same thing. Thanks