Google Antigravity has been free since its November 2025 launch. That alone makes it unusual. The AI IDE space is dominated by $20/month subscriptions, and Google walked in with a fully-featured agentic IDE powered by Gemini 3 Pro and offered it to anyone with a Gmail account. No credit card. No trial period. Just download and start delegating code to AI agents.

I’ve spent the past several weeks using Antigravity for real projects. Here’s the honest picture: it does things no other IDE currently does, and it has real problems you should know about before you trust it with your work.

Quick Verdict

Aspect	Rating
Overall Score	★★★★☆ (3.8/5)
Best For	Experimenters, vibe coders, personal projects
Pricing	Free (personal Gmail)
Parallel Agents	Excellent — up to 5 simultaneous
Model Quality	Excellent (Gemini 3 Pro)
Security	Poor — critical vulnerabilities unresolved
Enterprise Ready	No
Setup Ease	Very easy

Bottom line: Antigravity is the most technically ambitious free coding tool available. For personal projects and experimentation, it’s worth your time. For production work, proprietary code, or anything with credentials in the project folder, hold off until Google patches the security issues.

Try Google Antigravity →

What Google Antigravity Actually Is

Antigravity launched November 18, 2025, the same day Google released Gemini 3. It’s a fork of Visual Studio Code—same lineage as Cursor and Windsurf—but the philosophy is different.

Cursor and Windsurf assist you as you type. Antigravity treats the AI agent as the primary developer. You describe what you want built. The agent reads your codebase, plans the implementation, writes files, runs terminal commands, opens a browser, tests what it built, and fixes what’s broken. You supervise and approve.

That’s a meaningful shift. You’re not autocomplete-on-steroids territory anymore. You’re running a small software team where the developers happen to be AI agents.

The result is a tool that’s hard to evaluate by conventional benchmarks. It’s not “better” or “worse” than Cursor in the traditional sense—it’s doing something categorically different.

The Core Features Worth Understanding

Manager View: Parallel Agent Orchestration

This is Antigravity’s headline capability, and nothing else in the market matches it.

Manager View is a control panel that lets you spawn up to five AI agents working in parallel across separate workspaces. Agent 1 refactors the API layer. Agent 2 builds the frontend components. Agent 3 writes tests. Agent 4 handles documentation. All running simultaneously, none interrupting the others.

Each agent delivers Artifacts when it finishes: diffs, plans, screenshots, recordings. You review and approve before anything merges. The UI visualizes each agent’s status in real time.

I ran a three-agent session to build out a small personal project: one agent handling the database schema, one scaffolding the API endpoints, one wiring up the frontend. The experience is unlike anything I’ve had in another IDE. You stop being a developer in the traditional sense and start being a technical project manager.

The practical time savings are real. Three tasks that would have taken me several hours to cycle through sequentially were handled in parallel and ready for review in about 40 minutes.

Browser Agent: End-to-End Development in One Window

Antigravity’s browser sub-agent closes the feedback loop that other IDEs leave open.

The agent scaffolds a Next.js app in the editor, spins up the dev server, loads the running app in a built-in Chromium browser, runs end-to-end tests, notices a misaligned button, patches the CSS, verifies the fix, takes a screenshot, and hands you a summary. No tab-switching, no manual test runs.

For web development, this is a significant workflow improvement. The agent sees what the app actually looks like at runtime, not just what the code says it should look like.

Model Options

Antigravity isn’t locked to Google’s own models. It supports:

Model	Use Case
Gemini 3 Pro	Default — best for reasoning-heavy tasks
Gemini 3 Deep Think	Complex architecture and difficult problems
Gemini 3 Flash	Speed — fast completions and simple tasks
Claude Sonnet 4.5	Alternative reasoning model
Claude Opus 4.5	Most demanding tasks
GPT-OSS-120B	Open-weight option

The model flexibility matters. Claude handles certain reasoning patterns differently than Gemini. Being able to switch mid-session—Gemini 3 Pro for architecture decisions, Flash for rapid boilerplate—has real practical value.

Vibe Coding and Vision Input

Drop a screenshot of a Figma mockup into the chat. Antigravity’s vision capabilities (Gemini 3) read the layout, extract colors and spacing, and generate the CSS and JSX to replicate it.

For designers learning to code or developers working from visual specs, this eliminates a painful translation step. I gave it a screenshot of a dashboard I wanted to replicate and the output was about 85% correct on first pass—close enough that the remaining cleanup was faster than building from scratch.

The Rate Limit System

Antigravity’s limits work differently from typical API quota systems.

Limits aren’t calculated by requests or lines of code. They’re based on “work done.” A simple autocomplete uses almost nothing. Spawning three parallel agents on a complex task burns more. Rate limits refresh every five hours.

Personal Gmail accounts get generous limits for Gemini 3 Pro. Google AI Pro and Ultra subscribers get priority access with higher caps. Tab code completions are unlimited for all users.

In practice, I haven’t hit a hard wall during normal solo development sessions. Running multiple parallel agents on compute-intensive tasks occasionally approaches limits, but the five-hour refresh cycle makes it workable.

My Hands-On Experience

What Works Brilliantly

Parallel agents for independent work streams. When a project has separable tasks—database work, API work, frontend work—the Manager View delivers a productivity multiplier nothing else currently offers.

The browser integration. Web development with a browser agent that can see and react to what’s actually rendering is a meaningful leap over editor-only tools.

Gemini 3 Pro’s codebase understanding. Ask Antigravity “How does authentication work in this project?” and you get a walkthrough of your actual implementation, not a generic explanation of JWT. The codebase context is excellent.

Setup speed. Download from antigravity.google, drag to Applications, sign in with Gmail, open a folder. You’re running in under five minutes. The initial configuration screen walks you through choosing between Autopilot mode (agent runs terminal commands automatically) and Assisted mode (agent asks before executing). Choose Assisted mode.

Vision input for UI work. Not perfect, but good enough to be useful. Paste a design, get working code, clean it up.

What Doesn’t Work

Security is a genuine concern. I’ll cover this in full below, but it’s the biggest limitation. Real vulnerabilities exist. Google is working on them but hasn’t confirmed fixes.

Turbo mode is risky. Turbo mode lets agents execute terminal commands without confirmation. A documented incident in December 2025 saw a developer’s D: drive wiped because an agent executed a delete command against the wrong directory. The recycle bin was bypassed. The data couldn’t be recovered. Don’t use Turbo mode on machines with irreplaceable data.

No enterprise compliance story. No SOC 2. No SAML SSO. No audit trails. For professional developers at companies with security reviews, Antigravity doesn’t pass the requirements bar that Cursor and Windsurf both meet.

Agent errors require review. Agents make mistakes. They misunderstand scope, generate code that doesn’t integrate cleanly with existing patterns, or implement the letter rather than the spirit of a request. You need to review Artifacts carefully before accepting. This isn’t unique to Antigravity, but with five parallel agents producing output simultaneously, the review burden can pile up.

The Security Issues — Read This Before You Start

The security situation is real and documented. I’m not going to soft-pedal it.

Within 24 hours of launch, security researcher Aaron Portnoy documented five critical vulnerabilities with working exploits:

1. Remote code execution via indirect prompt injection
2. Hidden instructions embedded using invisible Unicode tag characters
3. No human-in-the-loop required for MCP tool calls
4. Data exfiltration via read_url_content
5. Data exfiltration via image rendering

The read_url_content exploit is the one to understand: a malicious comment or file in your project can trigger a prompt injection that reads your .env file and sends its contents to an external server. No dialog appears. No warning.

Google acknowledged the vulnerabilities and said teams were working on fixes. As of February 2026, no confirmed resolutions have been published.

Practical rules for safe Antigravity use:

• Don’t open untrusted or third-party code
• Don’t store API keys, secrets, or credentials in .env files within Antigravity projects
• Don’t use Turbo mode on production machines or drives with irreplaceable data
• Keep work projects and production credentials away from Antigravity sessions
• Use a dedicated machine or sandboxed environment if you want to test aggressively

This isn’t overcaution. These are reproducible, documented vulnerabilities. If your work involves client code, proprietary systems, or credentials, Antigravity isn’t the right tool yet.

Pricing

Plan	Cost	Details
Personal (Gmail)	Free	Gemini 3 Pro, generous rate limits, all features
Google AI Pro	$20/month	Higher rate limits, priority access
Google AI Ultra	$100/month	Highest rate limits, first access to new features
Enterprise	Not announced	No paid enterprise tier in current preview

The free tier is capable. I used it for three weeks of real project work without paying. Google is building the developer audience while the preview runs. Paid pricing for an Antigravity-specific tier hasn’t been announced.

Google Antigravity vs Cursor: The Honest Comparison

Feature	Antigravity	Cursor
Price	Free	$20/month Pro
Parallel agents	5 (Manager View)	1
Browser agent	Yes	No
Vision input	Yes	Yes
SOC 2 Type II	No	Yes
Model flexibility	Gemini 3, Claude, GPT-OSS	GPT-4, Claude, more
Turbo/auto-confirm	Yes (dangerous)	No
Enterprise compliance	None	Full
Track record	3 months	18+ months

If you’re a professional developer at a company with security requirements: Cursor. No contest.

If you’re experimenting, building personal projects, or want to explore what agentic parallel development actually feels like without paying: Antigravity.

For a broader look at how these tools stack up, see our Google Antigravity vs Cursor vs Windsurf comparison.

How to Get Started with Google Antigravity

1. Download from antigravity.google and select your OS (macOS, Windows, or Linux)
2. Install by dragging to Applications on Mac or running the installer on Windows
3. Launch and work through the initial setup wizard
4. Choose Assisted mode when prompted, not Autopilot or Turbo
5. Sign in with your Google account (personal Gmail is sufficient)
6. Open a folder, either an existing project or a new empty directory
7. Let it index — Antigravity reads your codebase before it can suggest or execute

For your first session, try a task with a narrow scope: “Add a contact form to this page” rather than “Build the entire frontend.” Evaluate the agent’s output quality on something you understand well before trusting it with complex work.

When you’re comfortable with single-agent results, open Manager View and try two parallel agents on separable tasks. That’s where the tool’s unique capabilities become apparent.

Who Should Use Google Antigravity

Personal projects and side work. No client credentials, no proprietary code, no stakes beyond your own time. This is Antigravity’s safe zone. The productivity gains from parallel agents are real and the free price is hard to argue with.

Developers exploring agentic workflows. If you want to understand where AI-assisted development is heading—multi-agent orchestration, browser-integrated feedback loops, vision-to-code workflows—Antigravity is the most accessible way to experience it.

Designers learning to code. Vision input (mockup screenshot → working UI code) is a more useful entry point for designers than traditional code completion.

Vibe coders and rapid prototypers. If you’re building MVPs, proof-of-concepts, or throwaway demos, the speed-to-working-code ratio is excellent. See our vibe coding platforms guide for the broader landscape.

Who Should Look Elsewhere

Professional developers at companies with security policies. Antigravity cannot clear a standard security review in its current state. Use Cursor or Windsurf.

Anyone storing credentials in their project. .env files in Antigravity projects are vulnerable to exfiltration by prompt injection. Don’t do it.

Teams requiring compliance. No SOC 2, no SAML, no audit trails, no enterprise support tier. See our best AI coding assistants guide for enterprise-appropriate options.

Developers who want stability. Antigravity is three months old. Bugs, changes, and rough edges are frequent. If you need a tool that’s predictable under production pressure, the maturity gap between Antigravity and Cursor is real.

The Bottom Line

Google Antigravity is the most technically interesting free coding tool released in years. Parallel agent orchestration with Manager View is ahead of anything Cursor or Windsurf currently offer. The browser integration closes a workflow gap that’s existed since AI IDEs first appeared. Gemini 3 Pro’s codebase understanding is excellent.

And it’s free. Right now, today, with your Gmail account.

The security issues are real and not yet resolved. That single fact determines who Antigravity is appropriate for in February 2026.

For personal projects, experimentation, and understanding where agentic development is heading: use it. Keep secrets off the machine. Use Assisted mode, not Turbo. Review every Artifact before accepting.

For professional work, client code, or anything with credentials: wait. Come back in six months. If Google patches the vulnerabilities and adds enterprise controls, Antigravity could reshape the market. The technical foundation is already better than what’s come before.

Download Google Antigravity →

---

Frequently Asked Questions

Is Google Antigravity free?

Yes, fully free for personal Gmail accounts. You get access to Gemini 3 Pro, all features including Manager View and browser integration, and generous rate limits that refresh every five hours. Google AI Pro ($20/month) and Ultra ($100/month) subscribers get higher rate limits and priority access, but the free tier handles personal use well.

How does Google Antigravity’s Manager View work?

Manager View is a control panel for spawning and monitoring multiple AI agents working in parallel. You open Manager View, assign tasks to separate agents (“Agent 1: refactor the API endpoints,” “Agent 2: build the user settings UI”), and they execute simultaneously in separate workspaces. Each agent produces Artifacts—diffs, plans, screenshots—for your review when it finishes. You approve what to keep. Nothing merges automatically.

Is Google Antigravity safe to use?

Not for professional or production work as of February 2026. Security researcher Aaron Portnoy documented five critical vulnerabilities within 24 hours of launch, including remote code execution and data exfiltration via prompt injection. A separate December 2025 incident saw a developer’s entire drive wiped by an agent running in Turbo mode. For personal projects on dedicated machines with no credentials stored in project folders, the risk is manageable with precautions.

What models does Google Antigravity support?

Antigravity runs Gemini 3 Pro (default), Gemini 3 Deep Think, Gemini 3 Flash, Claude Sonnet 4.5, Claude Opus 4.5, and GPT-OSS-120B. You can switch models within a session. The Gemini 3 models are included at no additional cost on all tiers.

How do rate limits work in Google Antigravity?

Rate limits are based on “work done” rather than number of requests. Simple tab completions use almost no quota. Running complex parallel agents on large tasks burns more. All quotas refresh every five hours. Tab code completions are unlimited regardless of tier.

How does Antigravity compare to Cursor?

Antigravity offers features Cursor doesn’t: parallel multi-agent orchestration, a browser sub-agent, and it’s free. Cursor offers what Antigravity doesn’t: SOC 2 Type II certification, 18+ months of production reliability, and multi-model flexibility for enterprise use. For personal projects, Antigravity is compelling. For professional work, Cursor is the safer choice.

When did Google Antigravity launch?

Google Antigravity was announced on November 18, 2025, alongside Gemini 3. It entered full public preview immediately after announcement and has been available to personal Gmail users since launch.

---

Last updated: February 21, 2026. Pricing and features verified against antigravity.google and the Google Developers Blog.