Gemini CLI and Claude Code do the same basic job. You open a terminal in your project folder, describe what you want in plain English, and the agent reads files, proposes edits and runs commands. The big differences are who makes them, which AI models they use, and how you extend them.
The short version
| Gemini CLI | Claude Code | |
|---|---|---|
| Made by | Anthropic | |
| What it is | Terminal agent | Terminal agent (plus IDE and desktop integrations) |
| Open source | Yes (Apache 2.0) | No |
| Models | Gemini models | Claude models |
| Free plan | Free access unclear (see below) | No. Needs a paid Claude plan or API credits |
| MCP servers | Yes, via ~/.gemini/settings.json | Yes |
| Project instructions | GEMINI.md (can be set to read AGENTS.md too) | CLAUDE.md, plus skills |
| Install | npm install -g @google/gemini-cli, Homebrew or npx | Native installer for macOS, Linux and Windows |
Prices and plans change. Check each vendor's own pages rather than trusting a table (including this one).
A note on cost
Gemini CLI used to be known for generous free use with a Google login. Its README now says free individual access and Google AI Pro / Ultra plans moved to Google's Antigravity CLI on 18 June 2026, which Google's Developers Blog confirms. See our Antigravity CLI getting started guide and Antigravity CLI vs Claude Code. Gemini CLI itself now covers Gemini API keys (usage-based billing), Vertex AI and Gemini Code Assist Standard / Enterprise licences. Google's quota page still lists a few free-tier limits, so the picture is a little unclear. If free use matters to you, check Google's current docs before you start. Older tutorials may describe the old setup.
Claude Code needs a paid Claude plan (Pro, Max, Ultra, Team or Enterprise), a Claude Console account with pay-as-you-go credits, or access through Amazon Bedrock, Google Cloud or Microsoft Foundry. The free claude.ai plan doesn't include it.
How they feel to use
Day to day they are very similar. You type a request, the agent works out which files matter, shows you what it wants to change and asks before it does anything risky. Both also work in a "headless" way, where you run a single prompt from a script instead of chatting. In Gemini CLI that looks like gemini -p "your prompt".
Gemini CLI also has a companion extension for VS Code that shows its edits as a diff you can accept or reject in the editor.
Where Gemini CLI is stronger
- Open source. You can read the code, and it's released under the Apache 2.0 licence.
- Easy to try.
npx @google/gemini-cliruns it without a permanent install. - Fits Google's world. It supports Vertex AI and Google Cloud licences, which can suit teams already on Google Cloud.
Where Claude Code is stronger
- More built-in ways to extend it. Skills, hooks, subagents and
CLAUDE.mdproject memory are first-class features. - Claude models. If you prefer how Claude writes and explains code, this is where you get it.
- A clearer paid path. The plans and credits are well documented on Anthropic's side.
What they share
Both support MCP (Model Context Protocol, a standard way to plug extra tools such as GitHub or a database into an AI). Both read a plain-text file in your project for standing instructions. Both ask permission before changing things by default.
Who should pick what
- You're on Google Cloud, or want an open-source agent: Gemini CLI.
- You want skills, hooks and subagents: Claude Code.
- You've never used a terminal: either will feel unfamiliar. Start with our Claude Code install guide, which walks through a first small change.
- You can't decide: try both on one small, real task. Because they work in the terminal, they don't conflict.
Whichever you choose, set up a safety net first. Our git safety net guide shows how to undo anything the AI gets wrong.
Our verdict
There's no universal winner. Gemini CLI is the open-source option from Google; Claude Code is the more feature-rich one with a clear paid path. See also Claude Code vs OpenAI Codex and Gemini CLI vs OpenAI Codex.