Google Antigravity explained: Antigravity is Google’s agentic development platform. AI agents plan programming tasks, write code, run terminal commands, and test web apps in a built-in browser. It consists of the Antigravity 2.0 desktop app, the Antigravity IDE, a command-line tool (CLI), and a Python SDK. It is free for individuals, with a weekly quota for agent work.
This guide covers what each component does, what Antigravity costs, how it compares with Cursor and Claude Code, and what businesses should check on data protection. It has nothing to do with the browser trick “Google Gravity” or with physics searches for “anti gravity”.
What’s New
Antigravity 2.0 has arrived: The desktop app runs several agents in parallel, schedules background tasks, and accepts voice commands.
The CLI and SDK: A new CLI built in Go and an SDK for custom agent building (The Next Web).
A new model: Google names Gemini 3.8 Flash as the engine of the platform and says it beats Gemini 3.1 Pro on coding and agent benchmarks.
New subscription tiers: Google added a $100-per-month AI Ultra tier with five times the Pro limits, and the top tier dropped from $250 to $200/month (The Next Web).
CLI Phasing Out: The legacy consumer Gemini CLI is officially being discontinued, and the Antigravity CLI GitHub repository has completely taken over its role.
Google Antigravity at a Glance
Item | Answer |
Developer | |
Introduced | November 2025, together with Gemini 3 |
Components | Antigravity 2.0 (desktop app), IDE, CLI, Python SDK, extensions for existing editors released on May 19, 2026 |
Models | Gemini ecosystems (including 3.8 Flash, 3.1 Pro, and Gemini 4 Argon) |
Price | Free for individuals with a weekly quota; higher capacity available with Google AI Pro ($19.99/mo) and Ultra ($ 99.99/mo) |
Platforms | Windows, macOS, Linux, and Googlebook |
Interface | English only |
What Is Google Antigravity?
Antigravity answers a simple question: what does software development look like when agents, not people, do most of the typing? You describe a goal in plain language, for example, “build a sign-up page with email validation and test it”. The agent writes a plan, builds the code, starts the app, clicks through the interface, and fixes errors.
The key idea is “agent-first”. A classic editor assistant reacts to what you type. Antigravity agents work on their own for longer stretches and show their work as artifacts: an implementation plan, a task list, screenshots, and test results. You can comment on plans and results without touching the code. For the wider background, see our article on generative AI vs agentic AI and on agentic AI in software development.
Antigravity 2.0, IDE, CLI and SDK: The Components
Component | What it does | Best for |
Antigravity 2.0 | Desktop app without a code editor, running several agents in parallel | Teams and anyone delegating tasks |
Antigravity IDE | Code editor with an agent side panel and tab completion | Developers who work in the code themselves |
Antigravity CLI (command agy) | Agents in the terminal | Terminal-first developers, automation |
Python SDK | Build your own agents on the same technology | Developers with custom workflows |
IDE extensions | Agent features inside existing editors | Anyone keeping their usual setup |
Antigravity 2.0 and the Manager
Antigravity 2.0 replaces the editor with an overview where you start, watch, and approve several agents at once. This is the Manager view: one agent writes tests, another refactors, a third checks the layout in the browser.
For multi-agent work in practice, read about multi-agent code review with Claude Code.
Antigravity IDE
The IDE is the classic editor with an agent side panel: tab completion, inline commands, and an agent for bigger tasks. If you prefer working in the code yourself, this is the one.
Antigravity CLI
The CLI brings the same agents to the terminal and replaces the Gemini CLI. For a primer on this tool class, see Gemini CLI vs Claude Code.
Key Features:
Autonomous agents: Plan tasks, write code, run commands, and adjust the plan.
Built-in browser: The agent launches your web app, fills in forms and checks the interface.
Artifacts: Plan, task list and screenshots show what the agent did.
Parallel agents: Several tasks run side by side in Antigravity 2.0.
Tab completion and commands: Unlimited on the free plan.
Scheduled tasks: Recurring work can be automated.
Safety levels: Depending on the setting, the agent asks before running commands or works alone.
Which Models Does Google Antigravity Use?
Antigravity runs on Google’s Gemini models, currently working with Gemini 3.5 Flash and Gemini 3.1 Pro. It has also been working with the newest additions to the Gemini family, that is Gemini 3.8 Flash and Gemini 4 Argon (which is explicitly available for Google AI Pro and Ultra subscribers). Depending on the plan, third-party models such as Anthropic’s Claude and the open GPT-OSS are also available. You cannot bring your own API key or endpoint (Data Studios). For the differences between the Gemini models, see Gemini 3 vs Gemini Pro vs Deep Think and our article on Gemini 3 Flash. The model list changes quickly, so check the current one in the app.
Google Antigravity Pricing: Is It Free?
Yes, for individuals. The free plan includes unlimited tab completions and commands. Agent work is limited by a quota that refills every five hours until a weekly ceiling is reached. More capacity comes with Google’s AI plans.
Plan | Price (USD) | What you get |
Individual | $0 | Weekly quota, unlimited tab completions, and terminal commands |
Google AI Pro | ~$20 / month | Higher limits; extra AI credits can be added |
Google AI Ultra | ~$100 / month | Expanded access with 5x the Pro limits |
Google AI Ultra (Max) | ~$200 / month | Maximum capacity with 20x the Pro limits |
Business | Usage-based | Central administration and custom spend caps via Google Cloud |
The official Antigravity pricing page is the authoritative source. Google doesn’t publish fixed request counts: limits follow how much work the agent does, not how many prompts you send. The free quota has been reduced several times in the past (Vibecoding.app). If you run agents heavily, expect to need a higher plan or extra credits.
How to Install Antigravity in 5 Steps
Check requirements: Windows 10 (64-bit), macOS 12 or later, or 64-bit Linux. It runs on Mac and Windows alike.
Download: Get the desktop app or the IDE from antigravity.google.
Sign in: With a personal Google account. Workspace accounts often need a personal Gmail address.
Create a project: Grant access to one folder or repository. The agent works only inside it.
Give a first task: Describe the goal, review the plan in the artifact, and approve it.
Data collection is on by default, so review the privacy settings right after installing.
Testing Antigravity: What to Look For
A short test with four tasks shows whether it fits your work:
Task | Example | What to watch |
Small web app | To-do list in React, saved locally | Time taken, number of corrections, does it run the first time? |
Debugging | A project with a deliberate bug | Is the cause found, is the plan clear? |
Browser test | Check a sign-in form | Quality of screenshots and error reports |
Limits | Work normally for a week | When does the weekly limit kick in? |
Antigravity vs Cursor vs Claude Code
Particulars | Google Antigravity | Cursor | Claude Code |
Free to use | Yes, weekly limit | Limited free tier | No, paid plan needed |
Entry price | Google AI Pro (about $20) | Pro (about $20) | Claude Pro (about $20) |
Bring your own key | No | Yes | Yes |
Models | Gemini, plus Claude and GPT-OSS depending on plan | Claude, GPT, Gemini and more | Claude |
Strength | Free entry, multi-agent app | Model choice, polished editor | Terminal-first, newest Claude models |
For more information, read our articles on Cursor Composer 2.5 and GPT-5.2 Codex vs Claude Code.
Pick Antigravity if you want a free start and several agents in parallel. Pick Cursor when model choice and a mature editor matter. Pick Claude Code for terminal work with the newest Claude models.
Antigravity Alternatives
Cursor: AI editor with wide model choice.
Claude Code: Anthropic’s terminal agent.
GitHub Copilot Workspace: Microsoft/GitHub's agentic environment for planning and building.
Local models: When code must not leave your machine (see Ollama vs LM Studio).
Security and Data Protection (GDPR) for Businesses
Agents with access to a terminal, files and a browser are powerful, and that makes them a risk. Clarify three points before rollout:
Approvals: Use the strict setting with approval steps until the agent works reliably in your project. In the unchecked mode it runs commands without asking.
Prompt injection: Agents read web pages and documents, and crafted content can slip instructions to them. See our guide to prompt injection.
Data handling: Data collection is on by default. For individual accounts, no EU data residency is documented and no data processing agreement is offered. For business code, access through Google Cloud or a local setup is the safer choice. This is not legal advice.
To roll out tools like this safely, our team offers AI consulting for businesses.
Who Should Use Google Antigravity?
Beginners and hobby developers: yes, thanks to the free plan.
Development teams: yes, if you want to delegate tasks in parallel. Test limits and costs first.
Companies with sensitive code: only with business access and a privacy review.
Anyone who wants fixed prices and their own API keys: Cursor or Claude Code fit better.
Conclusion
Google Antigravity is one of the strongest free entry points into agentic software development. The multi-agent app Antigravity 2.0 is the biggest difference from classic editors. The weak spots are opaque limits and open data protection questions for business use. To see which models sit behind tools like this, read our overview of the best AI reasoning models in 2026.
FAQ
What is Google Antigravity?
Google Antigravity is a platform where AI agents plan, run, and test programming tasks on their own. You set the goal and review the results.
Is Google Antigravity free?
For individuals, yes. The free plan has a quota for agent work. Tab completion and commands are unlimited.
How much does Google Antigravity cost?
$0 without a subscription. Google AI Pro is about $20 per month, and Ultra starts at about $100 per month. Businesses pay on a usage basis through Google Cloud.
What is the difference between Antigravity and Claude Code?
Antigravity offers a free start, a desktop app for several agents, and mainly Gemini models. Claude Code runs in the terminal, uses Claude models, and needs a paid plan or an API key.
What is the difference between Antigravity 2.0 and the Antigravity IDE?
Antigravity 2.0 is an app without an editor for managing several agents. The IDE is a code editor with an agent side panel.
Does Google Antigravity work on Mac and Windows?
Yes, on macOS 12 or later, Windows 10 (64-bit), and 64-bit Linux. There is no Android app.
Can I use Google Antigravity in German?
The interface is in English, but you can write to the agent in German and receive German answers.
Is Google Antigravity GDPR-compliant?
That cannot be confirmed in general. For individual accounts, no EU data residency is documented. Businesses should use access through Google Cloud and run a privacy review.
What is Google Gravity, and is it related?
No. “Google Gravity” is an old browser trick where the elements of the Google homepage fall to the bottom of the screen due to simulated gravity. Google Antigravity is an entirely unrelated, next-generation development platform.