🌏 中文版
Warning: Discontinuation Notice (2026/05/19) Google announced that Gemini CLI will cease service for all Free, Pro, and Ultra users on June 18, 2026. The successor is Antigravity CLI — Agent Skills, Hooks, and Subagents can all be carried over, while Extensions are renamed to Antigravity plugins with an automatic migration tool provided. Enterprise users (Gemini Code Assist Standard/Enterprise license or Google Cloud API key) are not affected for now. Migration docs: antigravity.google/docs/gcli-migration.
Gemini CLI is Google’s open-source terminal AI coding agent, with source code at github.com/google-gemini/gemini-cli. It brings Gemini models directly into your terminal, and its core selling point is straightforward: most developers never need to pay a cent.
This article breaks down Gemini CLI’s free tier design, three authentication methods, paid plans, core features, and important 2026 updates.
Product Positioning
Gemini CLI is Google’s answer to the “terminal Agent” space. Unlike Claude Code or GitHub Copilot CLI, it has been an open-source project from day one — anyone can fork, modify, and contribute. It’s backed by Google’s Gemini model family, with native integration into Google Search and the Vertex AI ecosystem.
In terms of positioning, it targets developers who want to use AI for development tasks in the terminal but don’t want to pay a monthly subscription. Google’s strategy is clear: attract users with an extremely generous free tier and make the Gemini ecosystem the default choice.
Free Tier: The Core Selling Point
Gemini CLI’s free tier design is the most aggressive in the industry. All you need is a Google account to get:
| Item | Free Tier |
|---|---|
| Request Rate | 60 requests / min |
| Daily Limit | 1,000 requests / day |
| Model | Gemini 2.5 Pro |
| Context Window | 1,000,000 tokens |
| Multimodal | Full support (images, code, documents) |
| Web Search | Built-in Google Search grounding |
| Built-in Tools | All features, no restrictions |
How did Google arrive at these numbers? They analyzed actual usage data from their internal developers, identified the heaviest users’ consumption levels, and then set the free limit at 2x that number. In other words, even Google’s most intensive internal developers would find the free tier more than sufficient.
The message is clear: the vast majority of developers will never hit the paywall. What you get isn’t a crippled version — it’s the full package with the strongest model (2.5 Pro), the largest context window (1M tokens), and all core features.
Three Authentication Methods
Gemini CLI supports three authentication paths, each corresponding to different use cases and billing models:
| Auth Method | Cost | Model | Context Window | Use Case |
|---|---|---|---|---|
| Google Account (Gemini Code Assist for Individuals) | Free | Gemini 2.5 Pro | 1M tokens | Individual developer daily use |
| Gemini API Key (Free Tier) | Free | Flash models only | Varies by plan | Programmatic calls, CI/CD |
| Vertex AI (Express Mode) | Free (no billing setup required) | Varies by quota | Varies by plan | GCP ecosystem integration |
Key differences between the three paths:
- Google Account is the recommended path. Just sign in with your Google account to automatically receive the full free tier listed above. The model family is automatically selected by the system (defaults to 2.5 Pro).
- Gemini API Key (Free Tier) is limited to Flash models only — no Pro. Suitable for scenarios requiring programmatic access without Google account authentication.
- Vertex AI Express Mode doesn’t require Google Cloud billing setup, making it ideal for teams already in the GCP ecosystem who want to run quick tests.
Paid Plans
When the free tier isn’t enough (though most people won’t hit that point), Gemini CLI offers these paid options:
| Plan | Cost | Description |
|---|---|---|
| Google AI Pro | $19.99/mo | 1-month free trial, increased usage limits |
| AI Ultra | $124.99 / 3 months (~$42/mo) | Highest limits, for heavy users |
| Workspace (Gemini Code Assist Subscription) | Via Google Cloud subscription | Enterprise team plan with admin features |
| Pay-as-you-go | API Key or Vertex AI usage-based billing | Per-token pricing, no caps |
Note: Google AI Plus (consumer subscription plan) currently does not apply to CLI API usage. If you’ve already subscribed to Google AI Plus, that quota won’t carry over to Gemini CLI. This is a common point of confusion.
The pay-as-you-go path is suitable for scenarios requiring high-volume Gemini calls in CI/CD or automation workflows. No daily limits — purely per-token billing.
Core Features
Gemini CLI’s feature set covers the major needs of a terminal Agent:
- 1M Token Context Window — This is the largest context window among all Agent CLIs currently available. Particularly valuable for large monorepos, allowing you to load an entire module’s codebase for analysis in one go.
- Multimodal Capabilities — Supports image input, enabling you to drop in screenshots for UI analysis, read charts, and understand design mockups.
- Google Search Grounding — Built-in Google Search lets the agent search the web in real-time to supplement its answers. No additional setup or payment required.
- File Operations — Read, create, and modify files; execute shell commands. Standard Agent toolset.
- Code Analysis — Understands code structure, traces call graphs, identifies patterns.
- Project Management Tools — Built-in tools for task tracking and progress management.
- Session Management — Persistent conversation history with the ability to switch between different sessions.
The 1M context window is the biggest differentiator. While Claude Code’s Opus offers deeper reasoning, its context window is much smaller. In scenarios requiring simultaneous comprehension of large numbers of files, Gemini CLI’s large context is irreplaceable.
Important 2026 Updates
Several notable changes in 2026:
- March 2026: Prepaid Billing Change — Google Cloud’s billing mechanism shifted to a prepaid model, affecting paid users on the Vertex AI path. If you’re using the Google Account free tier, you’re unaffected.
- June 1, 2026: Gemini 2.0 Flash-Lite Deprecation — If your workflow depends on the Flash-Lite model, you need to migrate before this date. Recommended alternatives are Flash or 2.5 Pro.
- Free Tier Bound to Google Cloud Project — The free limit is currently tied to the Google Cloud project level, not the individual API key level. This means all keys under the same project share the quota.
Use Cases
Gemini CLI is particularly well-suited for the following developers:
- Those who want a free, high-quality CLI Agent — The free tier covers 2.5 Pro and full functionality, which is sufficient for most individual developers. No credit card required, no subscription needed — just sign in with your Google account.
- Large Monorepo Developers — The 1M token context window lets you load massive amounts of code at once. If your project routinely exceeds hundreds of thousands of lines, this context size is a real productivity differentiator.
- Google / GCP Ecosystem Users — Teams already on Google Cloud can seamlessly integrate with Vertex AI, with permissions management and billing running through existing GCP infrastructure.
- Developers Who Value Open Source — Fully open source, forkable and customizable. Compared to the closed-source models of Claude Code or Copilot CLI, this is a decisive factor for some teams.
If what you need is the deepest reasoning capability (complex architecture design, multi-step debugging), Claude Code’s Opus remains the better choice. But if your primary needs are “free, good enough, and large enough context,” Gemini CLI is currently unmatched.
Migration to Antigravity CLI
After Gemini CLI is discontinued, the official recommendation is to migrate to Antigravity CLI — the terminal interface for Google’s Antigravity 2.0 platform. Rewritten in Go for faster execution, it also supports asynchronous background agents.
Migration Key Points
- Agent Skills: Global skills (
~/.gemini/skills/) are automatically read — no action needed - Extensions to Plugins: Run
agy plugin import geminifor automatic conversion - MCP Servers: Config file moves from
settings.jsontomcp_config.json; remote server field name changes fromurltoserverUrl - Quota Plans: Migrate to the corresponding Antigravity plan; see official migration docs for details
# Install Antigravity CLI (macOS / Linux)
curl -fsSL https://antigravity.google/install.sh | sh
# Migrate existing Extensions
agy plugin import gemini
Full migration guide: antigravity.google/docs/gcli-migration
Series Articles
This article is part of the Agent CLI series. For cross-tool comparisons of multi-model routing and subscription plans, see:
-> Agent CLI Subscription Plans and Multi-Model Routing Strategies
References
- Gemini CLI | GitHub
- Gemini CLI: Quotas and Pricing
- Gemini Developer API Pricing | Google AI for Developers
- Set up your coding assistant with Gemini MCP and Skills | Google AI
- Gemini Pricing in 2026 for Individuals, Orgs & Developers | Finout
- Google Developers Blog: Transitioning Gemini CLI to Antigravity CLI (Official Discontinuation Notice)
- Antigravity CLI Migration Docs: Migrating from Gemini CLI
Changelog
- 2026-05-21: Added Gemini CLI discontinuation notice (2026/06/18) and Antigravity CLI migration section; updated tldr, tags, and references
Loading...