Gemini CLI in 2026: Google's Free Terminal AI Agent vs Claude Code and Codex CLI

If you've been paying $20–$200/month for an AI coding assistant and wondering whether the price tag is justified, Google just dropped an answer: a terminal AI agent with 1,000 free requests per day, a 1-million-token context window, and zero credit card required.

That's Gemini CLI, which Google announced at I/O 2026. And for developers who already spend time in the terminal, it's worth understanding before you decide whether it earns a spot in your daily workflow.

What Gemini CLI Actually Is

Gemini CLI is Google's open-source terminal AI coding agent — structurally similar to Claude Code or OpenAI's Codex CLI, but built around Gemini 2.5 Pro. You open it inside a project directory, describe what you want in plain English, and it reads files, plans changes, proposes edits, runs shell commands, and asks you to approve each step.

The source is on GitHub under Apache-2.0, ships as a single npm package (@google/gemini-cli), and follows a weekly release cadence.

The headline differentiator is the free tier: authenticate with a personal Google account and you get 60 requests per minute and 1,000 requests per day against Gemini 2.5 Pro — with the full 1M-token context window. That's a complete day of real work for most developers, at no cost.

Getting It Running

Installation is straightforward on any platform:

npm install -g @google/gemini-cli
gemini --version

You'll need Node.js 20+. If you hit EACCES permission errors on Linux/macOS, switch to nvm rather than fighting the system npm — it's the right fix, not a workaround.

First run opens your browser for a Google account OAuth flow. Token gets stored in ~/.gemini/ and you're in. No API key needed for the free tier.

For automation, CI pipelines, or if you burn through the 1,000 daily limit, grab an API key from aistudio.google.com/apikey and set GEMINI_API_KEY in your environment. Billing then switches to per-token Google AI Studio rates with no daily cap.

Enterprise teams on Google Cloud use Vertex AI Application Default Credentials — set GOOGLE_APPLICATION_CREDENTIALS, GOOGLE_CLOUD_PROJECT, and GOOGLE_CLOUD_LOCATION, grant the service account the Vertex AI User role, done.

Day-to-Day Usage

Open a project directory and run gemini. You drop into an interactive prompt — describe what you need:

Add a Redis health-check endpoint to server.ts and cover it with a test in tests/health.spec.ts

The agent reads the relevant files, plans the changes, proposes diffs, and asks before touching anything. Each tool call — file read, file write, shell command — requires explicit confirmation by default.

Useful flags to know:

Built-in tooling is solid: file read/write/edit, shell execution, URL fetching, and Google Search grounding baked in. That last one is genuinely useful — the agent can pull live search results inline without you leaving the session.

For extending it, MCP support is there. Add servers to ~/.gemini/settings.json under mcpServers and they appear as tools in the model's context. Standard stdio MCP processes — same pattern as Claude Code's config.

Gemini CLI vs Claude Code vs OpenAI Codex CLI

All three are first-class terminal agents. Here's where they actually differ:

Benchmark performance. Claude Code (Opus 4.7) sits around 88% on SWE-bench Verified. OpenAI Codex CLI (GPT-5.5) is around 85%. Gemini CLI (Gemini 2.5 Pro) is around 76%. That ~12-point gap matters on hard multi-file refactors — first-pass correctness is noticeably better with Claude on complex changes. For everyday tasks, the gap compresses significantly.

Context window. Both Gemini CLI and Claude Code support 1M tokens. Codex CLI caps at 400K input / 128K output. If you're reasoning over a large codebase or a pile of long documents, Codex falls behind.

Price. This is where Gemini wins decisively. Claude Code requires a Pro ($20/mo) or Max ($100–200/mo) subscription. Codex CLI is pay-per-token with no free tier. Gemini CLI gives you 1,000 free requests daily. For solo developers or those evaluating whether a terminal agent fits their workflow at all, the free tier removes all friction.

Sandboxing. Codex CLI runs commands in a container by default — strongest isolation story. Claude Code and Gemini CLI both confirm per tool call. If you're running untrusted code regularly, Codex has the edge here.

License. Gemini CLI and Codex CLI are Apache-2.0. Claude Code is a proprietary CLI. If open source matters to your stack policy, that's a factor.

My take after using all three: if you're doing heavy multi-file feature work where you need the model to nail it on the first try, Claude Code is still the most reliable. If you're exploring a codebase, doing architecture planning, reading long documents, or just want a capable AI coding agent with zero subscription cost, Gemini CLI earns its place. I now use both — Gemini for exploration and planning sessions, Claude Code when correctness matters most.

What Works Well, What Doesn't

Works well:

Where it falls short:

Hardware That Levels Up the Terminal Workflow

A terminal AI agent is only as pleasant as the hardware it runs on. If you're spending significant time in a dev setup like this, it's worth the investment:

Should You Add It to Your Stack?

If you're already on Claude Code or Cursor and happy, Gemini CLI isn't a replacement — it's a complement. Use it for the long-context exploration and planning work where the free tier gives you room to iterate without watching your token budget. Switch to your paid tool when you need the accuracy on hard changes.

If you've been curious about terminal AI agents but not ready to pay a subscription to find out if they fit your workflow, Gemini CLI is the obvious starting point. 1,000 requests with Gemini 2.5 Pro and a 1M context window is enough to genuinely evaluate whether this style of development makes you faster.

For self-hosters running Proxmox and already managing Tailscale, n8n, and Ollama: Gemini CLI slots in cleanly. It runs locally, auth state lives in ~/.gemini, MCP config is portable, and the Vertex AI path integrates with existing GCP setups if you're there.

Install it, point it at a codebase, and see what happens.

npm install -g @google/gemini-cli
cd ~/your-project
gemini

---

Disclosure: This post contains affiliate links. If you purchase through these links, I may earn a small commission at no extra cost to you.