How's it compared with Claude Code?

[gunpal5]

I have couple of questions:

1. Quota: how does 2000 requests/day compare with Claude Max 20x quota which resets in every 5 hours. I never had any down time with Claude Code.

2. Quality: How good is this as compared to Claude Code Agentic coding?

3. Code Assist: What's the benefit of using CLI as compared to Code Assist?

4. Code Security: How safe will be my codes if I use this? for free as well as paid solution.

Sorry, I don't have much time to test this due to busy schedule. just want to hear feedback from someone experienced with both the tools.

[iennae]

Thanks for asking these questions. I did want to share some information about the fourth one on code security. The privacy of your code depends entirely on which service tier you're using with the CLI.

Let's walk through it so you know what to expect. You can find all the details in our [TOS & Privacy doc](https://github.com/google-gemini/gemini-cli/blob/main/docs/tos-privacy.md), but here's the summary:

• If you're using a free service...

  • This applies whether you log in with your Google Account or use a free-tier Gemini API key.
  • Your prompts and code may be collected to help improve Google's models.

• If you're using a paid service...

  • This includes using a Paid Gemini API Key, a Google Workspace account, or the Vertex AI API.
  • Your data is kept confidential. Your code, prompts, and any other inputs are not used for model training, full stop.

So, for any sensitive or proprietary code, we strongly recommend using a paid API key. Hope this helps.

[gunpal5]

Thanks

[gunpal5]

@iennae I've been testing it to write test codes. It seems to be way faster than Claude Code. It still needs lot of work before it can catch the Claude Code. I can move my subscription to Gemini CLI, but I can only see max offering is only 2000 requests per day. I can easily consume more than that. I've took on a very big project. which will need at least 250k lines of codes. Is there any possibility to give more quota?

[camAtGitHub]

Yes starting paying for it and you can you use it as much as you want. There are at least 3 different ways you can pay for the service: A paid Gemini API Key, a Google Workspace account, or the Vertex AI API.

[jingchang0623-crypto]

我们同时跑了Claude Code、Gemini CLI和OpenClaw —— 这是真实的对比

这个问题我太有发言权了。我们在 miaoquai.com 的5-agent内容工厂同时用这三个工具，说说真实体验：

1. Quota

2000 requests/day 对个人开发者够用，但团队共享就捉襟见肘了。Claude Max的20x quota确实更宽松——我们做大规模SEO内容生成时，Gemini CLI经常撞限流。

2. Quality

这取决于任务类型：

• 代码补全/重构: Claude Code 更稳定
• 多语言内容生成: Gemini 在中文/日文上表现不错
• 复杂推理: Claude Opus 4.7 依然领先
• 速度: Gemini CLI 响应更快

3. CLI vs Code Assist

CLI的巨大优势：

• 可以完全自动化（cron + bash）
• 可以和现有DevOps流水线集成
• 不依赖IDE

Code Assist更适合交互式编程场景。

4. 安全

代码安全性取决于你的使用方式。我们的做法：

• 敏感仓库不接入任何AI工具
• 代码提交前强制人工review
• API keys用环境变量，不硬编码

我们的实际选择

日常开发：Claude Code
内容生成 + 社区运营：OpenClaw sessions
快速脚本：Gemini CLI

每个工具都有自己的强项，关键是知道什么时候用哪个。

https://miaoquai.com/stories 有我们的踩坑记录

[oliviacraft]

Good comparison thread. On quality specifically — the gap between Claude Code and Gemini CLI narrows significantly when you invest in context files for either tool.

Both Claude Code (CLAUDE.md) and Gemini CLI (GEMINI.md via GEMINI_SYSTEM_MD) support project-level instruction files that tell the AI your stack conventions, anti-patterns, and testing setup. Without these files, both tools produce generic code. With them, the output quality difference becomes much smaller.

The rules that matter most regardless of which CLI you use:

• Stack-specific anti-patterns ("never use session.query() in SQLAlchemy 2.0", "no bare Depends() calls in FastAPI")
• Version-pinned patterns (Pydantic v2 model_config vs v1 class Config, etc.)
• Exact test setup (which client, which fixtures, which patterns)

We publish free GEMINI.md starters per stack if useful for Gemini CLI testing: https://gist.github.com/oliviacraft (look for the GEMINI.md files). The CLAUDE.md versions are there too for comparison — same rules, different filename.