Skip to main content
AI Tool Review Paid ★ 3.0/5

Gemini CLI

Terminal coding agent from Google with a 1M token context window. Consumer access ended June 2026. Enterprise and paid API access remain.

By The Codegen Team · Published June 23, 2026 · Updated June 2026

Visit Gemini CLI →
Pricing Paid
Rating 3.0/5
Setup time 5 minutes (npm install + API key or Google login)
IDEs Terminal CLI. VS Code (via Gemini Code Assist, enterprise only)

Gemini CLI is an open-source terminal coding agent built by Google under the Apache 2.0 license. Written in TypeScript, it launched in June 2025 and accumulated 104,000+ GitHub stars within a year.

Gemini CLI terminal session

Google shut down free, Pro, and Ultra access to Jules on June 18, 2026, redirecting consumer users to Antigravity CLI, a closed-source Go-based replacement.

Before the shutdown, the CLI was the only agentic coding tool running Gemini models at zero cost with the largest context window in the category. Google Search grounding let it pull live documentation and API references during execution without extra configuration.

A GEMINI.md file at the project root told the agent how the codebase was structured and which test commands to run. Same pattern OpenAI Codex uses with AGENTS.md.

The tradeoffs were real. Code output quality trailed the category leaders, and the agent drifted from instructions during longer sessions. Developers who needed precision used Claude Code for the hard problems and Gemini CLI for breadth.

Enterprise users on Gemini Code Assist licenses retain full access. Paid API authentication through Google AI Studio or Vertex AI still works at per-token rates.

Everyone else is looking at Antigravity CLI, Claude Code, or OpenAI Codex.

Key Features

1M token context window
1M tokens (Gemini 3 family)
Holds a full mid-sized repo in one session without trimming. Other tools needed paid tiers or special config to load the same amount of code.
Google Search grounding
Built-in
Fetches current web results during task execution without requiring an MCP server. Useful for verifying API docs and checking package versions mid-refactor without leaving the session.
GEMINI.md project instructions
Native
Sits in the repository root with test commands, coding conventions, and navigation hints. Nested files in subdirectories let teams apply different rules to different modules.
Open-source agent loop
Apache 2.0 (TypeScript)
The full ReAct loop, tool implementations, and sandboxing logic are public on GitHub. Settings live in ~/.gemini/settings.json, making the entire agent behavior auditable and forkable.
MCP server support
Via settings.json
Configure external tool servers in ~/.gemini/settings.json alongside the built-in file, shell, and search tools. The agent discovers and invokes MCP tools automatically based on the task.

Strengths & Limitations

Strengths
  • The context window held full mid-sized repos in one session, matching what others charged enterprise rates to unlock.
  • The only terminal agent with built-in web search, where others need MCP setup to pull live docs.
  • Over 6,000 community pull requests merged in under a year, with bug fixes moving from issue to release in days rather than months.
Limitations
  • SWE-bench Verified scores of roughly 76% trailed Claude Code at 88%, and multi-file refactors typically needed more iteration cycles to reach mergeable quality.
  • The agent did not follow instructions well. It changed code when told to analyze only, and longer sessions drifted from the original task.
  • Starting March 2026, Google restricted free-tier users to Flash models and prioritized enterprise traffic. Even Pro subscribers hit 429 rate-limit errors after two or three multi-turn prompts.

Who It’s For

Best for
Engineering teams on Google Cloud enterprise licenses who want terminal-native agentic coding tied to their existing infrastructure. Developers with paid API keys who need to hold a full monorepo in context for cross-file analysis.
Not ideal for
Individual developers without enterprise licenses or paid API keys. Consumer access is gone. For terminal agents, Claude Code offers the strongest output quality, OpenAI Codex runs in a cloud sandbox, and Antigravity CLI stays in the Google ecosystem.

Pricing Breakdown

Google announced the free tier shutdown at Google I/O 2026 and ended consumer access roughly 30 days later with no grace period. The old free plan offered 60 requests per minute and 1,000 per day.

Paid API access through Google AI Studio starts at $1.25 per million input tokens for Flash models. Pro models cost more. Enterprise contracts through Google Cloud offer custom rates and managed quotas.

Frequently Asked Questions

Comparisons featuring Gemini CLI

Build faster with AI-powered agents

See how Codegen automates the full development workflow — from ticket to pull request.

Get Started →