Gemini CLI

Learning Objectives

• Understand the completed June 18, 2026 sunset and the migration path to Antigravity CLI
• Understand how Gemini CLI works and what sets it apart from other coding agents
• Identify key features: model routing, plan mode, subagents, and MCP support
• Evaluate when to use Gemini CLI versus Claude Code, OpenAI Codex, or GitHub Copilot

What Is Gemini CLI?

Gemini CLI is Google's open-source command-line AI coding agent. It ran directly in your terminal with automatic model routing and a 1 million token context window. Google's launch materials paired it with Gemini 3.5 Pro for complex reasoning and Gemini 3.5 Flash for speed — though 3.5 Pro never shipped, so the Flash line carried the work in practice. The consumer-tier cutoff passed on June 18, 2026; new projects should target Antigravity CLI.

Unlike proprietary coding agents, Gemini CLI is fully open source (Apache 2.0) and available on GitHub. It is free to use with a generous daily limit tied to your Google account, making it an accessible entry point for developers who want to try agentic coding without a paid subscription.

Core Capabilities

Intelligent Model Routing

Gemini CLI analyzes each task and automatically selects the best model:

• Gemini 3.5 Pro: Named for complex reasoning and architecture work in Google's launch materials, but it never shipped — it remains partner-testing only as of August 2026
• Gemini 3.5 Flash: Used for quick questions, simple edits, code formatting, and routine operations

This routing happens transparently — you always interact with a single CLI, and it optimizes cost and speed behind the scenes.

Plan Mode

Plan mode lets you design implementation strategies before writing code. Gemini CLI creates a structured plan with steps, identifies files to modify, and gets your approval before executing. This is especially valuable for large features where you want to review the approach before any code changes.

Subagent Support

Gemini CLI can spawn subagents for parallel subtasks — delegating independent work items to separate agent instances that run concurrently. A model-driven parallel tool scheduler ensures safe tools execute simultaneously, improving throughput for multi-step workflows.

MCP Integration

Full support for the Model Context Protocol — connect Gemini CLI to databases (Supabase, Postgres), deployment platforms (Vercel, Firebase), project management tools (GitHub, Linear), and any other MCP-compatible service. Configure MCP servers in your project and Gemini CLI discovers their capabilities automatically.

gVisor Sandboxing

Gemini CLI uses native gVisor sandboxing (with experimental LXC support) to isolate agent actions from the host system. This provides safer execution of untrusted operations — particularly important when running agent-generated commands on your local machine.

JIT Context Loading

Just-In-Time context loading is enabled by default with deduplication — Gemini CLI loads only the context it needs for each step rather than loading your entire project upfront. This optimizes the 1 million token context window for maximum effectiveness.

Pricing

Gemini CLI is free with a daily usage limit:

• Free tier: Generous daily limit tied to your Google account — sufficient for most individual development workflows
• Paid accounts: Higher priority and expanded limits based on license type
• API key: Use your own Gemini API key for unlimited usage at per-token pricing

Strengths

• Free daily usage: No subscription required — start coding with AI agents at zero cost
• Open source: Full source code available; community contributions welcome; Apache 2.0 license
• 1 million token context: Among the largest context windows of any coding agent — handles very large codebases
• Automatic model routing: Optimizes between Pro (quality) and Flash (speed) without manual configuration
• Plan mode: Review and approve implementation strategies before any code changes
• Google ecosystem: Native integration with Google Cloud, Firebase, and Vertex AI

Limitations & Considerations

• Google account required: Free tier requires Google authentication
• Daily limits: Free usage is capped — heavy users may hit limits during intensive coding sessions
• Newer ecosystem: Fewer community extensions and integrations compared to Claude Code or GitHub Copilot
• Cloud inference: Code context is sent to Google's API — not suitable for air-gapped environments

Best Use Cases

When to choose alternatives:

• Maximum coding agent capability → Claude Code (Agent SDK, sub-agents, Agent Teams)
• GitHub-native workflow → GitHub Copilot (Issues, PRs, Actions integration)
• Browser-based development → OpenAI Codex (sandboxed cloud environment)
• Visual IDE experience → Cursor (AI-native editor with inline suggestions)

Migration to Antigravity CLI

The successor product is Antigravity CLI — a Go-based rewrite that ships alongside Google Antigravity 2.0 and replaces Gemini CLI as Google's terminal-side agentic coding entry point. The migration preserves the structural primitives developers depend on:

• Same model routing Google announced for Gemini CLI, though only the Flash side ever shipped
• Same Agent Skills, Hooks, Subagents, and Extensions — Extensions are renamed to "plugins" but the configuration shape is unchanged
• New: async workflows for complex multi-agent tasks that previously had to run sequentially
• New: unified backend shared with the Antigravity desktop app, so terminal and IDE sessions can hand off in-flight work
• New: faster execution courtesy of the Go runtime; Google highlights responsiveness as the headline improvement

Who needed to act: Google AI Pro, Ultra, and free-tier users — that cutoff has now passed, so install Antigravity CLI if you have not already. Enterprise customers using paid API keys for Gemini are unaffected — Gemini CLI continues to serve those requests past the cutoff. Migration documentation and walkthrough videos are linked from the Google Developers Blog announcement at developers.googleblog.com.

Key Takeaways

• Gemini CLI reached end of life on June 18, 2026 for Google AI Pro, Ultra, and free-tier users — Antigravity CLI is the supported replacement
• Enterprise customers on paid API keys are unaffected by the sunset; Gemini CLI continues to serve those requests
• Antigravity CLI preserves the same Agent Skills, Hooks, Subagents, and Extensions (renamed plugins) plus Gemini 3.5 model routing, and adds async workflows and a unified backend with the Antigravity desktop app
• Before the sunset, Gemini CLI was Google's open-source coding agent with automatic model routing across the Gemini Flash line
• 1 million token context window, plan mode, subagent support, and MCP integration make it a capable production tool through June 18, 2026
• Open source (Apache 2.0) under the legacy product; the Antigravity CLI source is also published on the Google Antigravity GitHub org