Claude Code is great, but alternatives exist with unique strengths. I tried a few.
TL;DR: the "coding agents" space is extremely dynamic with fierce competition. We can expect a lot of changes in the coming months, from form factor to mental models. Competition is good! Trying other coding agents may help you:
- stay open-minded about the UX / form factor
- increase your performance by using their unique strengths
- save credits: they have free tiers 💰
Here is a short list, in order of personal preference/relevance:
⭐️⭐️⭐️ Codex by OpenAI
"The same agent everywhere you build"
- ✅ generous free tier
- ✅ frontier models
- ✅ built-in worktrees and cloud environments
- 💡 not a CLI! Codex is primarily an app (though they do have a TUI version)
- 💡 integrated browser that the agent can access, with annotation tool
Overall the strongest contender. Integrated browser is extremely useful (all coding agents can access the browser, but some setup is required). Integrated agent orchestration + all-in-one app makes a great UX.
⭐️⭐️ openCode
"The open source AI coding agent"
- ✅ free tier (not frontier models) + low-cost plans
- ✅ Can connect to any model = you can use frontier ones
- ✅ Open Source (extensible / tweakable)
- ✅ Enterprise plan with confidentiality promises
- ✅ has an app like Codex
- 💡 killer feature seems to be the plans and open-source nature
⭐️ Gemini CLI by Google
"Build debug & deploy with AI"
- ✅ generous free tier
- ✅ frontier models
- ✅ must be decent and reliable since it's Google
- ✅ procurement may be easier since it's Google
- normal TUI coding agent, I didn't notice any original feature (typical Google?)
Note to Google: make your landing page scrollable and immediately list your USPs
Build, debug & deploy with AI | Gemini CLI
geminicli.com
Vibe by Mistral AI
"Agentic coding that meets you where you work. Write, test, and deploy autonomously with full codebase context."
- ✅ free tier
- ✅ European company
- ✅ decent
- 💡 killer feature seems to be sovereignty since it's a rare non-USA one.
- still has room to grow: I encountered a backend error during first use
Amp
"the frontier coding agent built for leading models, and what comes next"
recommended by an opinion leader (maybe Steve Yegge?), I wish I'd tried it, unfortunately Amp discontinued their free tier: https://ampcode.com/news/amp-free-is-ad-free
Did I miss a notable one? Any notable Chinese coding agents worth trying? Feel free to share in the comments.
Top comments (4)
Solid roundup. A few things worth adding to the conversation.
Cursor still deserves a mention here. It sits in a slightly different category since it’s a full IDE rather than a CLI or agent layer, but the agentic features it’s been shipping lately put it squarely in this space. A lot of teams are using it as their primary coding agent without even thinking of it that way.
On the Chinese agents question, Baidu’s Comate and Alibaba’s Tongyi Lingma are worth looking at, especially if you’re working with teams in that region. They’re not widely discussed in Western dev circles but they’re maturing fast.
The free tier point is underrated. Most developers don’t switch tools because of features, they switch because they hit a credit wall mid-project. Amp losing their free tier is probably the single biggest reason they’ll lose mindshare to Codex and openCode right now, regardless of quality.
One thing I’d add to your evaluation criteria is how well each agent handles multi-file context. Single file edits are solved. Where these tools diverge fast is when you’re refactoring across a large codebase. That’s the real stress test.
Curious whether you tried any of these on a real project or mostly evaluation tasks? The gap between demo performance and production use is still pretty wide for most of them.
Thanks for this very thorough addition.
I definitely need to re-try Cursor, which I dismissed as a "last gen" AI assistant. I'll try Google Antigravity as well.
Codex is good indeed.
Did you test Gas Town?
not yet but I heard about it.