Gemini CLI in 2026: Who It Is Still For After Antigravity

7 min read · Last updated: October 24, 2026

Gemini CLI is Google's open-source AI agent for the terminal, licensed Apache 2.0. As of June 18, 2026, its README says unpaid tier and Google One users were moved to Antigravity CLI, and that Gemini CLI is not optimized for the latest Gemini models. It still ships weekly and still works with API keys, Vertex AI and paid Code Assist licenses.

Came here from Instagram, LinkedIn, X or YouTube? Here is the full source list you came for. Everything below is built from these, nothing else.

The repo and README

• google-gemini/gemini-cli on GitHub, README read October 24, 2026

Key takeaways

• Unpaid tier and Google One users had Gemini CLI replaced by Antigravity CLI on June 18, 2026.
• The README says Gemini CLI is not optimized for Gemini 4 Argon and points you to Antigravity for it.
• The project is still fully open source under Apache 2.0 and ships on 3 release channels: nightly, weekly preview, weekly stable.
• 3 documented auth paths remain: Google sign-in, a Gemini API key, and Vertex AI.
• Headless mode with --output-format json and stream-json is the strongest reason to keep it in your scripts.

What happened to Gemini CLI?

The top of the Gemini CLI README now carries two notes. The first says that for unpaid tier and Google One users, Gemini CLI was replaced by Antigravity CLI on June 18th, 2026. The second says Gemini CLI has not been optimized for Gemini 4 Argon, and that you should use Antigravity for that model.

A third note repeats the message in the main section and again under Getting Started: "Gemini CLI is not optimized for the latest Gemini models. For the best coding experience with Gemini models use Antigravity CLI." The repo itself is not archived. It still lists weekly releases, a roadmap and an open contribution process.

So Google has not killed the tool. It has moved the free crowd and the newest model to a different product, and left Gemini CLI as the open-source, lightweight option.

Why does this matter to a developer?

It matters because the free path most people used to try Gemini CLI is gone, and the README itself tells you it is no longer the best Gemini coding tool. If you signed in with a personal Google account and no paid plan, your default terminal agent is now Antigravity CLI, not this repo.

There are three groups to think about.

You used it free with a Google account. Per the README, that path now leads to Antigravity CLI. Your old habits, like typing gemini in a project folder, need a new home.

You use an API key, Vertex AI or a paid Code Assist license. Nothing in the README says these stopped working. All three are still documented with exact environment variables.

You built scripts on it. The headless flags, the MCP server config and the GEMINI.md context files are all still documented. Your automation is not broken by this note, but you are now on a tool Google itself calls "not optimized" for its newest models.

That last point is the real trade-off. You keep an Apache 2.0 codebase you can read, fork and pin. You give up being on the tool Google tunes first.

Is Gemini CLI still free and open source?

Gemini CLI is still fully open source under the Apache License 2.0, and you can install it from npm, Homebrew, MacPorts or conda at no cost. What changed is access through the unpaid Google tier: those users were moved to Antigravity CLI. Running it now means bringing an API key, Vertex AI, or a paid organization license.

The README describes the tool as giving "the most direct path from your prompt to our model." That is still the pitch: a thin agent between your terminal and Gemini, with no IDE required.

If you want to compare where Gemini sits against other assistants before you pay for any of this, our ChatGPT vs Claude vs Gemini vs Grok breakdown is the place to start.

How do you install Gemini CLI?

You can run Gemini CLI without installing anything using npx @google/gemini-cli, or install it globally with npm install -g @google/gemini-cli. Homebrew and MacPorts packages exist for macOS, Homebrew also covers Linux, and the README gives an Anaconda route for restricted environments where you cannot install Node globally.

# No install
npx @google/gemini-cli

# Global install
npm install -g @google/gemini-cli

# Homebrew (macOS/Linux)
brew install gemini-cli

# MacPorts (macOS)
sudo port install gemini-cli

For locked-down machines, the README uses conda to get Node.js first:

conda create -y -n gemini_env -c conda-forge nodejs
conda activate gemini_env
npm install -g @google/gemini-cli

Which release channel should you use?

Pick stable unless you are testing. Here is the schedule exactly as the README states it.

Channel	Tag	When it ships	What it contains
Nightly	@nightly	Every day, 00:00 UTC	Everything on main, pending validations
Preview	@preview	Tuesdays, 23:59 UTC	Not fully vetted, may have regressions
Stable	@latest	Tuesdays, 20:00 UTC	Last week's preview plus bug fixes

The useful detail: stable is always last week's preview, promoted. If you hit a bug in preview on a Wednesday, the fix has a week to land before it reaches stable.

What changes in practice for authentication?

In practice, you now need to choose one of three authentication paths on purpose. Google sign-in works with a paid Code Assist license if you set your Cloud project. A Gemini API key gives you model choice and usage-based billing. Vertex AI is the path the README recommends for enterprise teams and production workloads.

Option	Best for (per README)	What you set
Sign in with Google	Paid Code Assist license from your organization	GOOGLE_CLOUD_PROJECT
Gemini API key	Specific model control, paid tier access	GEMINI_API_KEY
Vertex AI	Enterprise teams, production workloads	GOOGLE_API_KEY + GOOGLE_GENAI_USE_VERTEXAI=true

The commands:

# Option 1: Google sign-in with an org Code Assist license
export GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID"
gemini

# Option 2: Gemini API key (from aistudio.google.com/apikey)
export GEMINI_API_KEY="YOUR_API_KEY"
gemini

# Option 3: Vertex AI
export GOOGLE_API_KEY="YOUR_API_KEY"
export GOOGLE_GENAI_USE_VERTEXAI=true
gemini

The API key route is the one most solo developers will land on. The README lists two benefits: you can choose specific Gemini models, and you can upgrade for higher limits when you need them. Vertex AI adds what the README calls advanced security and compliance, higher rate limits with a billing account, and integration with existing Google Cloud infrastructure.

What can Gemini CLI still do well?

Gemini CLI is still a solid scriptable agent. It can query and edit large codebases, build apps from PDFs, images or sketches, and debug with natural language. Its best feature for automation is non-interactive mode, which returns plain text, structured JSON, or a stream of newline-delimited JSON events you can monitor.

Here is the headless mode, straight from the README:

# Plain text answer
gemini -p "Explain the architecture of this codebase"

# Structured output for scripts
gemini -p "Explain the architecture of this codebase" --output-format json

# Real-time events for long-running jobs
gemini -p "Run tests and deploy" --output-format stream-json

That stream-json flag is the one to notice. It gives you newline-delimited JSON events, so a CI job or a dashboard can watch a long run instead of waiting for one final blob.

The other features the docs still cover:

• MCP (Model Context Protocol) servers, configured in ~/.gemini/settings.json, so you can type things like @github List my open pull requests.
• GEMINI.md context files for persistent project context.
• Checkpointing to save and resume conversations.
• Token caching to cut token usage.
• Sandboxing and trusted folders to control what it can execute and where.
• Multiple directories with --include-directories ../lib,../docs.
• Model pinning with -m, for example gemini -m gemini-2.5-flash.

If your work is mostly scripts and pipelines, pinning a model and a CLI version is a feature, not a weakness. Agents that wire into other tools are a growing category; we covered another one in what Stripe Projects CLI does for AI agents.

Should you switch to Antigravity CLI?

Switch to Antigravity CLI if you were on the unpaid tier, use a Google One plan, or want Gemini 4 Argon, because the README says Gemini CLI is not optimized for it. Stay on Gemini CLI if you depend on API-key or Vertex AI billing, need an Apache 2.0 codebase, or run it headless in scripts that already work.

A quick way to decide:

Your situation	Recommendation
Free Google account, no paid plan	Antigravity CLI (Google already moved you)
Google One subscriber	Antigravity CLI (named in the README note)
Want the newest model, Gemini 4 Argon	Antigravity CLI
API key or Vertex AI billing	Gemini CLI still documented and working
Headless CI scripts, pinned versions	Gemini CLI
Want to read or fork the source	Gemini CLI (Apache 2.0)

This article only covers what the Gemini CLI README says. I am not going to describe Antigravity CLI's features, because the source here does not.

What should you do this week?

This week, check which auth path you are actually on, pin a release channel, and move any free-tier habits to Antigravity CLI. If you keep Gemini CLI for scripts, pin the model with -m and switch your automation to --output-format json so a change in output text does not break your parsing.

1. Check your auth. Run env | grep -E 'GEMINI|GOOGLE' and see which of the three paths you use. No variables and a personal account means you were on the path that moved.
2. Pin your channel. Install @google/gemini-cli@latest on any shared or CI machine. Leave @nightly and @preview for your own laptop.
3. Pin the model. Add -m to every script so a default change does not surprise you.
4. Use structured output. Swap plain -p calls for --output-format json in pipelines, and stream-json for anything long-running.
5. Write a GEMINI.md. Persistent context is cheap to add and works today.
6. Report bugs from inside the tool. The README says /bug files issues directly from the CLI.

If you want to run Google models with no account at all, our guide to running Gemma 4 locally for free is the other route.

FAQ

Is Gemini CLI discontinued?

No. The repo is still active, open source under Apache 2.0, and ships nightly, weekly preview and weekly stable releases. What changed on June 18, 2026 is that unpaid tier and Google One users were moved to Antigravity CLI.

Can I still use Gemini CLI with an API key?

Yes. The README still documents export GEMINI_API_KEY="YOUR_API_KEY" followed by gemini. It lists model selection and usage-based billing as the benefits of this option.

Does Gemini CLI support Gemini 4 Argon?

The README says Gemini CLI has not been optimized for Gemini 4 Argon, and tells you to use Antigravity for that model. For older models you can still pick one with the -m flag.

How do I use Gemini CLI in a script?

Use non-interactive mode with gemini -p "your prompt". Add --output-format json for structured output, or --output-format stream-json for newline-delimited JSON events during long runs.

Pin your Gemini CLI version and model today, then try our free tools on Art of Code or see how the Gemini 3.7 Flash speed trade-off plays out.

Written by Adithya Guttha, Founder of Art of Code.