Google Antigravity 2.0 (2026 Guide)

What it is: Google Antigravity is Google’s agentic development platform. It launched as an agent-first IDE (a fork of VS Code) in November 2025 alongside Gemini 3, and with version 2.0 at Google I/O 2026 it became a standalone platform with a desktop app, CLI, and SDK for running teams of agents in parallel. Free to start.
Who it is for: Developers who want to delegate longer coding tasks to autonomous agents and verify the results with screenshots and recordings rather than reviewing every line.
Best if: You want to try the most ambitious agent-first IDE on the market, plus you want to use Claude Sonnet 4.6 or Opus 4.6 alongside Gemini 3 Pro under one roof.
Skip if: You prefer a terminal-first workflow, Claude Code is the right tool. For one issue per day on what’s shipping, subscribe to our free daily newsletter.

What is the bottom line on Google Antigravity?

Google Antigravity is Google’s agent-first IDE, a VS Code fork that prioritizes autonomous agents over single-cursor coding. Released November 18, 2025 alongside Gemini 3 and currently in free public preview, it ships with two main views: an Editor view (the familiar IDE) and a Manager view that orchestrates up to 5 parallel agents. Agents generate “Artifacts”, task lists, screenshots, browser recordings, so you can verify their work without reading every diff. Native model support includes Gemini 3 Pro, Gemini 3 Flash, GPT-OSS-120B (the open-weights OpenAI variant), and notably Anthropic’s Claude Sonnet 4.6 and Opus 4.6. Available on macOS, Windows, and Linux.

What are the key takeaways?

• Released November 18, 2025 alongside Gemini 3 announcement.
• VS Code fork: familiar editor surface, totally new agent layer on top.
• Up to 5 parallel agents in the Manager view.
• Built-in Chrome browser for visual verification of UI work.
• Multi-model: Gemini 3 Pro, Gemini 3 Flash, Claude Sonnet 4.6, Claude Opus 4.6, GPT-OSS-120B.
• Free public preview as of May 2026. Paid tier coming.
• Cross-platform: macOS, Windows, Linux.
• Agents generate Artifacts: verifiable deliverables (plans, screenshots, recordings).

What changed in Antigravity 2.0 (Google I/O 2026)?

Google relaunched Antigravity at I/O 2026 in May, and the shift is big enough to change how you think about it. The first version, released in November 2025, was an agent-first IDE built as a fork of VS Code. Antigravity 2.0 moves the focus away from the code editor and toward managing teams of AI agents. The editor is still there; it is just no longer the point.

• A revamped desktop app built around multi-agent orchestration. You run several agents in parallel, spin up subagents to split a job, and schedule tasks to run in the background.
• A new Antigravity CLI, built in Go, that replaces the Gemini CLI. Google is retiring consumer access to the Gemini CLI and Gemini Code Assist IDE extensions on June 18, 2026. The Antigravity CLI keeps the features people relied on: Agent Skills, Hooks, Subagents, and Extensions, now called Antigravity plugins.
• An SDK for building your own agents, plus managed execution and enterprise support.
• Voice commands, so you can talk to your agents instead of typing every instruction.

The short version: Antigravity started as an IDE with agents bolted in. Version 2.0 is an agentic platform that happens to include an editor. The rest of this guide covers how it works, which models it runs, and how it compares to the alternatives.

What is Google Antigravity exactly?

Google describes Antigravity as an “agentic development platform” rather than an IDE per se. The underlying editor is a heavily modified fork of VS Code, familiar enough that any VS Code user can sit down and start working, but the value proposition is the layer on top: autonomous agents that read, write, and run code, with you supervising rather than driving.

The product has two primary views:

• Editor view: the usual IDE surface with an agent sidebar, conceptually similar to Cursor or GitHub Copilot.
• Manager view: a control center that orchestrates multiple agents working in parallel across workspaces, supporting asynchronous task execution.

Manager view is where Antigravity’s “agent-first” thesis lives. Instead of one agent helping you write one file, you launch five agents simultaneously on five tasks and check back when each finishes.

How does Antigravity compare to Cursor and Claude Code?

Dimension	Antigravity	Cursor	Claude Code
Primary mode	Manager view + Editor view	IDE with AI sidebar	Terminal CLI + IDE plugins
Built on	VS Code fork	VS Code fork	Native CLI, IDE plugins
Agent parallelism	Up to 5 parallel	Sequential	Agent teams (Opus 4.6+)
Built-in browser	Yes (Chrome)	No	No (use Playwright MCP)
Models	Gemini 3, Claude 4.6, GPT-OSS	Claude, GPT, Gemini, others	Sonnet 4.6 + Opus 4.7
Pricing	Free preview (paid tier pending)	$20/mo (Pro)	$17–$20/mo (Claude Pro)
Maker	Google	Cursor / Anysphere	Anthropic
Best for	Parallel autonomous tasks	Interactive AI coding	Terminal + agent automation

If you want to try the most ambitious agent-orchestration experience available today, Antigravity is the answer. If you want a polished AI-in-the-editor experience, Cursor. If you want a terminal-first workflow with deep IDE integration, Claude Code.

Which AI models does Antigravity support?

• Gemini 3 Pro: default. Google’s flagship multimodal model.
• Gemini 3 Flash: faster, cheaper. Right for routine work.
• Claude Sonnet 4.6: Anthropic’s mid-tier model, natively supported.
• Claude Opus 4.6: Anthropic’s flagship public model, natively supported.
• GPT-OSS-120B: the open-weights variant of OpenAI’s models.

Cross-model support is a strong differentiator. You can run a Gemini 3 Pro agent on planning, a Claude Opus 4.6 agent on the actual code change, and a Gemini 3 Flash agent on test-writing, all in parallel. Each model brings its own strengths; Antigravity lets you assemble them in one workflow.

What are Antigravity Artifacts and why do they matter?

Artifacts are Antigravity’s solution to the trust problem in agentic coding. Instead of just producing raw tool calls (file edits, shell commands, API responses), agents produce verifiable deliverables:

• Task lists: what the agent plans to do, before doing it.
• Implementation plans: structured outline of changes the agent intends to make.
• Screenshots: via the built-in Chrome browser, the agent can demonstrate the result of UI changes.
• Browser recordings: for verifying end-to-end flows.
• Diff summaries: human-readable description of what changed.

The thesis: when an agent finishes a task, you should be able to verify the result without re-reading every diff line. Artifacts give you the lightweight verification surface.

How does Antigravity run 5 parallel agents?

The Manager view is the control center. You queue tasks, each becomes an agent, and the manager runs them in parallel, up to five at a time. Tasks can span multiple workspaces; agents can coordinate through shared artifacts.

Practical workflow: kick off five concurrent improvement tasks for your codebase (“add error handling to the API layer,” “update README with new install steps,” “write tests for the recently-changed payment module,” “audit dependencies for known CVEs,” “draft a migration plan to the new auth library”). Walk away. Each agent works independently and posts its Artifacts when done. You review the five outputs together rather than five at a time over five hours.

This is the same pattern Claude Code’s Opus 4.6 agent teams enable, but Antigravity exposes it as a first-class workflow rather than a per-task option.

How do you install Google Antigravity?

1. Go to antigravity.google
2. Click “Get started” and pick your platform: macOS, Windows, or Linux
3. Download the installer (similar to downloading VS Code)
4. Open the installer and follow the prompts, standard application install on every platform
5. Launch Antigravity
6. Sign in with your Google account (required for the free preview)
7. Pick a project folder to open and start using it like VS Code, or open the Manager view to launch parallel agents

The free preview includes “generous rate limits on Gemini 3 Pro” per Google’s announcement. Claude model access requires you to provide your own Anthropic API key inside Antigravity’s settings. OpenAI GPT-OSS-120B is available through the open-weights option.

How much does Antigravity cost?

As of May 2026, Antigravity is free, in public preview, with full Gemini 3 Pro access and no announced usage caps. Google has indicated a paid tier will follow once the preview ends, but no pricing has been published.

Caveats on “free”:

• Claude models cost extra. Using Sonnet 4.6 or Opus 4.6 in Antigravity routes through your own Anthropic API key, billed at Anthropic’s standard pay-as-you-go rates.
• GPT-OSS may use compute. If you run GPT-OSS-120B locally or on a hosted endpoint, that compute cost is yours.
• Future paid tier expected. Once Antigravity exits preview, expect a per-seat or per-usage Antigravity-branded plan, separate from Gemini 3 API pricing.

How well does Antigravity work with Claude?

Surprisingly well, given Antigravity is a Google product. Claude Sonnet 4.6 and Opus 4.6 are first-class supported models in Antigravity, not bolted-on extensions but native model options in the agent picker. In practice this means:

• You can use Antigravity’s Manager view to orchestrate 5 parallel Claude agents.
• You can mix Claude and Gemini agents in the same workspace, one agent on Opus 4.6 for hard reasoning, another on Gemini 3 Flash for routine work.
• Claude’s strong long-context performance pairs well with Antigravity’s parallel-agent model since each agent can hold a large slice of the codebase in mind without losing thread.

The strategic implication: Google chose to make Antigravity open to competing models rather than Gemini-only. This is similar to how Cursor supports Claude, GPT, and Gemini natively. The era of single-model IDEs is ending; agent-first IDEs are pluggable by default.

Should you switch from Claude Code, Cursor, or Copilot to Antigravity?

Three scenarios where switching makes sense:

• You want to test agent-first workflows. Manager view is the most polished parallel-agent experience available. If you’ve been curious about delegating coding tasks to multiple agents at once, Antigravity is the cleanest place to try it.
• You use Gemini 3 Pro and want full integration. Antigravity is the Gemini-native development environment. If Gemini is your primary model, this is the path of least friction.
• You want multi-model parallelism in one workspace. Cursor supports multiple models in sequential agent calls; Antigravity’s Manager view runs them simultaneously. For complex projects this is a meaningful workflow difference.

Three scenarios where you shouldn’t switch:

• You’re a terminal-first developer. Antigravity is GUI-first, stick with Claude Code.
• You’re invested in the Cursor ecosystem. Cursor’s polish and ecosystem maturity (extensions, settings, community) is years ahead of Antigravity’s preview.
• You don’t want a paid tier surprise. Antigravity is free during preview, but the paid tier is coming. If you bake critical workflows around it, plan for that transition.

Frequently asked questions

Is Google Antigravity free?
Currently yes, in public preview as of May 2026. A paid tier is expected once preview ends. Using Claude models requires your own Anthropic API key, which is billed separately.

Is Antigravity better than Cursor?
Different strengths. Antigravity’s Manager view is the strongest parallel-agent UX on the market. Cursor’s editor experience and ecosystem maturity are still ahead. For pure agent-first work, Antigravity. For polished AI-in-editor work, Cursor.

Can I use my VS Code extensions in Antigravity?
Yes, Antigravity is a VS Code fork, so most VS Code extensions work out of the box. The exception is extensions that conflict with Antigravity’s agent layer; those will either be disabled or warned about on install.

Does Antigravity work offline?
The editor surface works offline (it’s VS Code under the hood). Agent functionality requires internet because the models are API-hosted.

How do I use Claude Sonnet 4.6 in Antigravity?
Open Antigravity settings, find the model providers section, paste your Anthropic API key, and select Sonnet 4.6 (or Opus 4.6) as the default model for an agent. Usage routes through your Anthropic account, billed at standard API rates.

Can Antigravity be used in CI/CD?
Antigravity is GUI-first and not designed for headless CI/CD use. For programmatic agentic workflows in CI, use the Claude Agent SDK or Google’s Gemini API directly.

Is Antigravity Open Source?
No, Antigravity itself is proprietary Google software. It’s a fork of the open-source VS Code base, but the agent layer and Manager view are closed source.

How does Antigravity differ from Project IDX?
Project IDX was Google’s earlier cloud-based development environment. Antigravity replaces and supersedes it as Google’s flagship dev tool. IDX users are being migrated to Antigravity.

What’s the relationship between Antigravity and Gemini 3?
Antigravity launched alongside Gemini 3 (November 18, 2025) as the showcase IDE for Google’s new model family. Gemini 3 Pro is the default model in Antigravity, but Antigravity also supports Claude and OpenAI models.

Can I run Antigravity on Chromebook / ChromeOS?
Officially supported through ChromeOS’s Linux container. Performance varies with hardware; lower-end Chromebooks may struggle with multiple parallel agents.

Does Antigravity work with private GitHub repositories?
Yes. Agents have the same access to your local filesystem and Git remotes that VS Code does. For private GitHub repos, just clone the repo and Antigravity’s agents can read, edit, branch, commit, and push using your existing Git credentials. The built-in Chrome browser also signs into GitHub.com if you let it.

How does Antigravity handle long-running agentic tasks?
The Manager view is designed for asynchronous task execution, you can queue tasks and walk away. Agents that take longer than the active session stay running on Google’s infrastructure (for Gemini agents) or your local machine (for Claude/GPT-OSS agents using your own API key). You receive notifications when each task completes.

What data does Antigravity send back to Google?
Per Google’s documentation, Antigravity sends model inputs and tool calls to the Gemini API endpoints. For Claude or OpenAI models, requests route to those providers’ APIs instead. Google maintains usage telemetry for the product itself. For enterprise use, contact Google to discuss data-residency and audit-log options.

Is Antigravity good for Python data-science work?
Reasonable, with caveats. Notebooks (.ipynb files) open as VS Code notebooks; agents can read and edit them. The Manager view is overkill for typical notebook work but useful for batch-processing operations across many notebooks at once. For dedicated data-science workflows, JupyterLab + Claude Code still has the edge.

How does Antigravity handle terminal commands?
Agents can run shell commands via a built-in terminal tool. You can configure per-task permission gates, some agents may be approved to run any command, others restricted to read-only operations. This is similar to Claude Code’s permission modes but exposed through the GUI.

Will Antigravity work with my existing CLAUDE.md files?
Yes for the Claude-model side: when an Antigravity agent uses Claude Sonnet 4.6 or Opus 4.6, it picks up your project’s CLAUDE.md just as Claude Code would. This is a meaningful interoperability win, the same project context file works across both products.

Track every Antigravity update in one daily email

Google ships meaningful Antigravity updates regularly, new agent capabilities, new model integrations, paid-tier announcements. The Beginners in AI newsletter ships one issue every day covering what’s new across Antigravity, Claude Code, Cursor, and the rest of the agent-IDE space. Free, daily, no fluff.

Sources

• Google Antigravity homepage
• Build with Google Antigravity, Google Developers Blog
• Getting Started with Google Antigravity, Google Codelabs
• Google I/O 2026 keynote: Antigravity coverage
• Google Antigravity overview
• Claude Opus 4.6 announcement, Anthropic