Claude Code vs Gemini CLI

Free is a real argument. It's just not the only one.

Gemini CLI launched in mid-2025 and the headline was hard to ignore: Gemini 2.5 Pro in your terminal, open source, no credit card required. 60 requests per minute, 1,000 per day on the free tier. Google wasn't messing around.

I tested it. It's genuinely capable for one-off tasks. The model is good. But after a few days I understood why I still reach for Claude Code when the work is real.

Where Claude Code wins

Configuration depth. CLAUDE.md is genuinely unmatched. You write your project rules once — stack preferences, coding style, file conventions — and every session respects them. Gemini CLI has no equivalent. Every session starts from zero context about your preferences. When you tell Claude Code "we use next-auth v5, not passport," it remembers that. Gemini starts fresh every time.

Persistent memory across sessions. Your team's naming conventions, that weird deployment quirk, the API you said never to call directly. Claude Code carries that forward. I've had it remind me of a decision I made six weeks ago. Gemini CLI doesn't have this at all.

Hooks and extensibility. Pre-tool hooks, post-tool hooks, custom skills, MCP servers. Claude Code's extension surface is massive. You can make it format code before every commit, run linters after edits, or connect to any API through MCP. Gemini CLI is barebones here. It's a capable agent; it's not a platform.

Sub-agents. Claude Code can spin up parallel sub-agents to divide work. Research in one thread, implementation in another, tests in a third. Gemini CLI runs single-threaded.

Tool use reliability. Claude Code's file edits and terminal commands are battle-tested across millions of sessions. It rarely botches a file. Gemini CLI is newer and rougher around the edges. I've seen it write to the wrong path more than once.

Where Gemini CLI wins

Price: it's free. 60 requests per minute, 1,000 per day. For hobby projects, learning, or exploration, that's a serious advantage. Claude Code costs $20/month minimum.

Context window. Gemini 2.5 Pro handles up to 1M tokens. If you need to dump an entire large codebase into context and ask questions about it, Gemini has raw capacity Claude can't match. Claude Opus tops out at 200K.

Google ecosystem. If you're deep in Google Cloud, Firebase, or Vertex AI, Gemini CLI integrates more naturally with those services. Simple as that.

Open source. Gemini CLI is fully open source. Claude Code's CLI is source-available but not open source in the traditional sense. If that distinction matters to you, Gemini wins here.

My actual take

Gemini CLI is great for exploration and one-off tasks. Exploring a new repo, doing a quick refactor, learning to code without paying for anything. For that use case, it's excellent and I'd genuinely recommend it.

But for daily professional use — working on the same codebase day after day, where conventions matter, where you want the AI to get smarter about your project over time — Claude Code's configuration system pulls away hard. The combination of CLAUDE.md, persistent memory, hooks, and MCP extensibility creates something Gemini CLI doesn't have: compounding context. The longer you use it on a project, the more effective it gets. That's worth $20/month to me.

The best move is to try both. Gemini CLI costs nothing to test. But once you've built a solid CLAUDE.md with memory and hooks behind it, you'll understand the difference.

When to use which

Use Gemini CLI when you're exploring a new repo, doing quick one-off tasks, learning to code, or don't want to pay for anything yet. It's a solid free tool and I mean that sincerely.

Use Claude Code when you're working professionally on a codebase you'll touch every day. When conventions matter. When you want the AI to carry context forward instead of starting from scratch every session.