Google Antigravity Tutorial — Run Gemini 3 + Claude in One Free IDE (2026)

If you've wanted to try Claude Opus 4.6 or Sonnet 4.6 on real coding tasks but don't want to pay Anthropic for a subscription, this is the easiest path I've seen: Google Antigravity — a free AI coding IDE that puts Claude, Gemini 3, and GPT-OSS 120B in one model dropdown, all billed through your Google account.

In this tutorial we'll cover:

• What Antigravity is and how it differs from Cursor / Claude Code
• Install on macOS / Windows / Linux
• The six models on the free tier (and which one is actually slow)
• The Agent Manager — multi-agent workflow with built-in git worktree isolation
• How to use Claude inside Antigravity without an Anthropic account
• One-click MCP installs (Cloud Run, Firebase, BigQuery, AlloyDB, etc.)
• A full feature comparison with Claude Code

What Is Google Antigravity?

Antigravity is Google's agent-first development platform. Google launched it in November 2025 alongside Gemini 3.

Three things to know up front:

• It's a standalone IDE — not a CLI like Claude Code, not a VS Code extension. Its own application that runs on macOS, Windows, and Linux.
• It has two views: an Editor view (file tree + code + agent sidebar — familiar territory if you've used Cursor) and a Manager view (a workspace for orchestrating multiple agents in parallel).
• It's free for individuals during the public preview — no credit card required.

The Manager view is the part Google emphasizes, and once you start using it you understand why. More on that below.

Install Antigravity

1. Go to antigravity.google/download.
2. Pick your platform — macOS, Windows, or Linux.
3. Download the installer, drop the app into Applications, and launch it.
4. Sign in with your Google account.

No API key. No credit card.

That's the whole install.

The Free Tier — Six Models, Per-Model Weekly Quotas

Once signed in, the free tier gives you six models in the dropdown:

• Gemini 3.1 Pro (High reasoning)
• Gemini 3.1 Pro (Low reasoning)
• Gemini 3 Flash
• Claude Sonnet 4.6 (Thinking mode)
• Claude Opus 4.6 (Thinking mode)
• GPT-OSS 120B (Medium)

Each model has a separate weekly quota. You can see the bars in Settings → Models with a Refreshes in 6 days, 23 hours label. They reset every seven days.

One important warning from real testing

On the free tier today, Gemini 3.1 Pro at the High setting takes a long time to first response — sometimes minutes for a one-line answer. Use it for hard reasoning tasks where you can wait, but for day-to-day coding the practical free-tier models are:

• Claude Sonnet 4.6 (Thinking)
• Claude Opus 4.6 (Thinking)
• GPT-OSS 120B (Medium)
• Gemini 3 Flash

These all responded in 15–30 seconds during our test session.

The Editor View

This is what you'd expect from a modern AI IDE: file tree on the left, code in the middle, the Agent panel on the right. If you've used Cursor or VS Code with Copilot, you're already at home.

Two shortcuts to remember:

• Cmd + L → focuses the Agent panel
• Cmd + E → switches to the Agent Manager

The Agent Manager is the view that's actually different from a normal IDE. That's where Antigravity's pitch lives.

Agent Manager — Multi-Agent Workflow with Worktree Isolation

Press Cmd + E and Antigravity opens the Agent Manager in its own window.

Left sidebar lists your Workspaces — every folder you've ever opened in Antigravity is in there. Pick one, and any new conversation you start in that workspace becomes its own agent, running in parallel with the others.

Each conversation also has an environment toggle:

• Local — runs in your current workspace
• Worktree — spins up a fresh git worktree so parallel agents don't step on each other

That worktree toggle is genuinely useful. With Claude Code you'd manually git worktree add to isolate parallel agents. Antigravity does it for you per-conversation.

Demo: three agents on a Next.js project

In the video I spin up three conversations in one workspace:

1. Agent 1 → Build a UserCard React component in components/UserCard.tsx with name, email, avatarUrl props using Tailwind.
2. Agent 2 → Write Jest tests for UserCard covering rendering, props, and the empty-avatarUrl fallback.
3. Agent 3 → Add a .card-hover-glow CSS rule with a subtle blue shadow and document it in styles/README.md.

Three conversations, three agents, all running at the same time on the same workspace. Click any conversation in the sidebar to watch its agent work — tool calls, file edits, terminal output. When the agent finishes, that conversation pane shows the completed file changes.

Each conversation is your review surface. You see exactly what each agent did, ask it to change something, or open the file directly to inspect the change.

Use Claude Without an Anthropic Account

This is the part most viewers miss — and the single biggest reason to install Antigravity if you've been priced out of Claude.

Open the model dropdown. Claude Sonnet 4.6 and Claude Opus 4.6 are sitting right there next to the Gemini models. Both run in Thinking mode by default.

You don't need:

• An Anthropic API key
• A Claude Pro subscription
• A Claude Max subscription

Billing for Claude inside Antigravity flows through Google.

So if you've ever wanted to try Claude Opus 4.6 on a real coding task without paying Anthropic for it, this is the easiest path I've seen. Pick "Claude Opus 4.6 (Thinking)" in the dropdown, send your prompt, and Opus runs the agent loop for you — same Manager workflow, just a different brain.

One caveat — Custom models

There's a Custom model option at the bottom of the dropdown for bringing your own endpoint. The tooltip on hover reads:

Custom models are only supported on Cloudtop / Linux machines.

Translation: on macOS or Windows, the six built-in models are what you get. No way to point Antigravity at your own Anthropic-compatible endpoint unless you're on Cloudtop or Linux. If you need that flexibility on Mac, Claude Code is still the answer.

MCP Support — One-Click Installs for Google Cloud

Open Settings → Customizations. Antigravity ships with one-click MCP installs for the Google ecosystem:

• Cloud Run
• Google Kubernetes Engine (OSS)
• Dart
• Firebase
• Genkit
• BigQuery
• AlloyDB for PostgreSQL
• Spanner
• Cloud SQL for PostgreSQL / MySQL / SQL Server

There's also an Open MCP Config button at the top to wire in any custom MCP server — same Model Context Protocol as Claude Code, so existing MCP servers (filesystem, GitHub, browser, etc.) work the same way here.

If you're already invested in Google Cloud, the one-click installs save real setup time.

Antigravity vs Claude Code — Honest Comparison

My take

Both tools have parallel agents, MCP, diffs, and undo. The split is mostly cost, form factor, and how you get to Claude.

• If you don't have an Anthropic subscription and you want Claude → Antigravity is the cheaper path, because billing flows through Google.
• If you live in your terminal, want Plan mode, want the rewind checkpoint flow, or you need to point at a custom Anthropic-compatible endpoint on macOS → Claude Code is still the better fit. Claude Code also currently has access to Opus 4.7, which Antigravity doesn't yet expose.

The two aren't mutually exclusive. Install both. Use whichever fits the task.

Resources

• 🔗 Download Antigravity: antigravity.google/download
• 📺 Full video tutorial: youtu.be/W1gLEoCqMbg
• 📖 Google Developers Blog: Build with Google Antigravity
• 🧪 Codelab: Getting Started with Antigravity
• 📜 Antigravity Docs: Review Changes in Manager

Related videos on AyyazTech

• Run Gemma 4 31B with Claude Code (Ollama Cloud)
• Run DeepSeek V4 in Claude Code
• Run Claude Code 95% Cheaper (OpenCode + MiniMax M2.7)

Final Word

If Anthropic's pricing has been blocking you from trying Claude on real coding work, Antigravity removes that block — for free, today. Install it, sign in with Google, pick Opus 4.6 from the dropdown, and ship something.

If you want a follow-up where I build a real production app live in Antigravity — full backend, full deploy — drop a comment on the video and I'll script it next.

Subscribe to AyyazTech on YouTube — I cover every AI coding tool worth your time so you don't waste yours.