Google Antigravity Explained: Features, Pricing & Setup Guide (2026)

If you've checked your site analytics lately, you've probably noticed a strange new term climbing the charts: Google Antigravity. It shows up in search suggestions, developer forums, and YouTube thumbnails, but most explanations either assume you already know what an "agentic IDE" is or bury the practical details under marketing language. This guide fixes that. It's the deep, current explainer we wished existed when we started researching this platform — covering what Antigravity actually is, how its features work, what it costs in 2026, how it compares to Cursor and Windsurf, and how to set up AGENTS.md files the right way.

What Is Google Antigravity?

Google Antigravity is Google's agentic development platform — a tool that lets you hand off software tasks to autonomous AI agents instead of writing every line yourself. Google unveiled it on November 18, 2025, alongside the launch of Gemini 3, and opened it immediately to the public as a free preview.

Structurally, Antigravity is built on a heavily modified fork of Visual Studio Code (some in the community argue it's closer to a fork of Windsurf, which is itself VS Code-based, but Google has not confirmed the exact lineage). That means the editor itself will feel instantly familiar if you've used VS Code, Cursor, or Windsurf before. What's different is the philosophy underneath it.

Most AI coding tools to date have been "editor-centric": you type, the AI suggests a completion or answers a question in a side panel, and you stay the one driving every keystroke. Antigravity flips that relationship. It's "agent-centric" — you describe a goal in plain language, and an autonomous agent plans the work, writes the code, runs it in a terminal, and even opens a browser to visually verify the result, largely without you supervising every step. Google frames this as building trust, autonomy, feedback, and self-improvement directly into the product, rather than treating AI as a bolt-on feature of a traditional editor.

The Antigravity Product Family in 2026

One thing that trips up a lot of people searching for "Google Antigravity release date" or "Antigravity documentation" is that Antigravity isn't a single app anymore — it's grown into an ecosystem. As of the Google I/O 2026 announcements in May, here's what exists under the Antigravity name:

• Antigravity IDE — the original, full-featured agentic code editor that launched in November 2025. Still actively recommended for developers who want a deeply integrated coding experience.
• Antigravity 2.0 — a new standalone desktop application announced at I/O 2026. Unlike the original Agent Manager, which lived inside the IDE, Antigravity 2.0 runs independently as a command center for orchestrating multiple agents across projects, even outside of a single codebase.
• Antigravity CLI — a lightweight, terminal-only way to spin up agents without opening a graphical interface at all, aimed at developers who prefer staying in the command line.
• Antigravity SDK — for teams that want to build their own custom tooling or products on top of Antigravity's agent orchestration layer.

If you only need one of these, the IDE is still the most common starting point and the one most tutorials assume you're using.

Google Antigravity Features: A Full Breakdown

1. The Editor View

This is the part of Antigravity that looks and behaves like a conventional code editor. You get tab completions, inline AI commands, and a familiar file tree and terminal layout. It exists for moments when you want to be hands-on — fixing a specific line, reviewing a diff, or making a small tweak yourself rather than delegating it.

2. The Manager Surface (Agent Manager / Mission Control)

This is Antigravity's signature feature, and the one most queries about "agent manager features" or "mission control" are really asking about. Instead of the agent living inside your editor, the Manager Surface flips that: it's a dedicated interface for spawning, observing, and orchestrating multiple agents that work asynchronously, often across different projects at once. You can kick off a background research task in one workspace while actively coding in another, and the Manager's inbox notifies you as agents make progress, hit a decision point, or finish.

3. Browser Control and Autonomous Verification

Agents in Antigravity aren't limited to writing code. They can launch a local server, open an embedded Chrome browser, click through the actual UI they just built, and confirm it behaves as expected — closing a loop that previously required a human to manually test every change.

4. Artifacts

To solve the "trust gap" of AI-generated work — where you have no way to verify an agent's claim that something is "fixed" without re-reading all the code — Antigravity generates Artifacts as it works. These include structured task lists, implementation plans for you to review before code changes happen, screenshots, browser recordings, and code diffs. You can comment directly on these artifacts (similar to commenting in Google Docs) to redirect an agent mid-task.

5. Model Optionality

Antigravity isn't locked to a single AI provider. As of 2026, it supports Gemini 3.1 Pro and Gemini 3.5 Flash as the default models, alongside Anthropic's Claude Sonnet 4.6 and Claude Opus 4.6, and an open-weight GPT-OSS-120B model from OpenAI. This matters in practice: Flash is fast and cheap for routine work, Gemini Pro and Claude Sonnet handle architecture and reasoning-heavy tasks well, and Claude Opus tends to produce the strongest results on the hardest problems — at the cost of burning through usage quota fastest.

6. Skills

Skills are reusable, on-demand instruction packages. Each skill lives in its own folder containing a required SKILL.md file (plus optional scripts, references, and assets) that describes a specific task — like enforcing a commit message format or following a particular database migration process. Skills aren't loaded into the agent's context all the time; the agent scans available skill descriptions at the start of a conversation and only pulls in the full instructions when a task actually matches. This keeps the agent focused and avoids cluttering its context window with irrelevant rules.

What's New in Antigravity 2.0 (I/O 2026)

Antigravity 2.0, announced on May 19, 2026, expanded the platform considerably beyond its November 2025 launch state. The headline additions include:

• Dynamic sub-agents — the primary agent can now break a large task into smaller pieces and spin up helper agents to work on them in parallel, speeding up large-scale refactors and migrations.
• Scheduled tasks — you can configure an agent to run on a recurring schedule in the background, for example checking build health every morning or generating a status report without you triggering it manually.
• Ecosystem integrations — native connections across Google AI Studio, Android, and Firebase, including a one-click "Export to Antigravity" that brings an entire project (with its full context) from AI Studio into local development.
• Antigravity in the Gemini Enterprise Agent Platform — lets Google Cloud customers connect Antigravity directly to their cloud projects for enterprise workflows.
• Native Android app generation — building Android apps from a prompt, with a direct path to publishing through the Google Play Console.

How Much Does Google Antigravity Cost in 2026?

This is one of the most search-worthy and most misunderstood parts of the platform, so it's worth being precise. Antigravity launched in November 2025 as a free public preview with what Google described as generous rate limits. That changed considerably over the following months, and the pricing structure has shifted more than once.

As of mid-2026, here's the breakdown:

• Free tier — still available, includes access to all supported models (Gemini, Claude, GPT-OSS) but with limited rate caps that refresh roughly every five hours. No credit card required, but realistically this tier functions more like a trial than a daily-driver setup.
• Google AI Pro — $19.99/month — higher quota than free, generally pitched at hobbyists, students, and developers who spend more time in the editor than delegating to agents.
• Google AI Ultra — $100/month — introduced at I/O 2026, aimed at developers and technical leads who need regular but not constant agent usage. Includes a 5x higher usage limit than the Pro plan, priority access to Antigravity, and bundled perks like 20TB of cloud storage and YouTube Premium.
• Google AI Ultra Max — $200/month (cut from $249.99 at I/O 2026) — the highest quota tier, intended for developers using Antigravity as a primary daily tool.

In March 2026, Google also introduced an AI credits system as a pay-as-you-go overage mechanism once your plan's included quota runs out, priced around $25 for 2,500 credits, though Google has not publicly disclosed exactly how many tokens or requests one credit converts to. Worth knowing before you commit: developer communities on Reddit and in coverage from outlets like The Register and DevClass have documented repeated rate-limit reductions since launch, with some Pro-tier users reporting their original quotas cut by over 90% within a few months. If you plan to rely on Antigravity heavily, it's worth checking the current limits on Google's official pricing page before subscribing, since this is an area that has changed faster than most documentation can keep up with.

Google Antigravity vs Cursor vs Windsurf vs Codex

If you're choosing between AI coding tools in 2026, here's how Antigravity's positioning differs from the alternatives most often compared against it.

Tool	Core Model	Best For	Starting Price
Google Antigravity	Agent-centric (Manager Surface)	Delegating multi-step, end-to-end tasks with autonomous verification	Free tier; Pro $19.99/mo
Cursor	Editor-centric, with strong agent mode	Developers who want hands-on control with AI assistance close by	Free tier; Pro ~$20/mo
Windsurf	Editor-centric, "Cascade" agent flow	Similar audience to Cursor, with its own proprietary model options	Free tier; paid plans available
OpenAI Codex	Agent/CLI hybrid	Developers already in the ChatGPT/OpenAI ecosystem wanting higher rate limits	Bundled with ChatGPT plans; Pro tier $100/mo

The practical takeaway: Antigravity's biggest differentiator is the Manager Surface and its Artifacts system, which make it genuinely better suited to "set it running and check back later" workflows than tools that expect you to babysit every suggestion. Cursor and Windsurf still tend to win for developers who want to stay closely involved in every change. Codex has become a serious rate-limit alternative for teams frustrated by Antigravity's quota changes.

How to Get Started With Google Antigravity

Setup is straightforward, and you don't need anything beyond a Google account and a Chrome browser (used for the agent's browser-testing capability):

1. Download Antigravity from antigravity.google for macOS, Windows, or Linux.
2. Sign in with a personal Gmail account when prompted.
3. Choose a model — Gemini 3.1 Pro is the default, but you can switch to Claude or GPT-OSS depending on the task.
4. Select a development mode. Antigravity offers three: Agent-driven development ("autopilot," where the agent writes, runs, and tests code with minimal check-ins), Review-driven development (the agent asks permission before nearly every action), and Agent-assisted development (the recommended middle ground, where you stay in control but the agent handles safe automations on its own).
5. Open or create a project folder, go to the Agent Manager panel, and click "New Task" to describe what you want built.

For your first real task, keep the scope narrow. A common beginner mistake is asking an agent to build an entire application in one prompt; you'll get better, more reviewable results by breaking work into focused, single-purpose tasks and reviewing the implementation plan before the agent starts writing code.

AGENTS.md, GEMINI.md, and Skills: Best Practices for Antigravity in 2026

If you're going to use Antigravity seriously, configuring its rule files properly is what separates a smooth agentic workflow from a frustrating one. Here's how the system is structured and how to use it well.

Rule file hierarchy

• Global rules live in your home directory at ~/.gemini/GEMINI.md (Antigravity-specific) and increasingly ~/.gemini/AGENTS.md, a newer cross-tool standard also recognized by Cursor, Windsurf, and other agentic editors. Use these for personal preferences that should apply to every project — language choices, documentation conventions, or hard rules like never hardcoding API keys.
• Workspace rules live inside the project directory itself (GEMINI.md or AGENTS.md at the project root) and define team-specific standards, like which framework or state management library to default to.

Practical best practices

• Define clear boundaries. Use your AGENTS.md to specify exactly which files or directories each agent is allowed to modify, especially once you're running more than one agent on the same codebase.
• Start small and scale up. Begin with two or three agents working on isolated features or bug fixes before expanding to larger parallel workflows.
• Add human checkpoints for anything risky. Require explicit approval for database migrations, third-party integrations, or anything touching production infrastructure.
• Maintain a GUARDRAILS.md. When an agent makes a mistake or gets stuck in a repetitive loop, document the failure pattern in a dedicated guardrails file so future tasks avoid the same trap.
• Use Skills for repeatable, codified processes. Anything you find yourself explaining to an agent more than once — a deployment checklist, a code review standard, a commit message format — belongs in a SKILL.md file rather than a one-off prompt.
• Be careful with terminal and MCP permissions. Since agents can have real terminal access, use read-only credentials wherever possible when connecting MCP servers (for databases, GitHub, etc.), and review your terminal auto-execution policy rather than leaving everything on full autopilot by default.

Common Challenges and Limitations

Antigravity is genuinely capable, but it isn't without friction points worth knowing before you build a workflow around it:

• Quota unpredictability. As covered above, rate limits and the newer credit system have changed multiple times since launch, and Google's own documentation notes that published limits aren't guaranteed.
• Requires internet access. Since the core models run in the cloud, Antigravity isn't usable offline.
• A real learning curve. Developers used to writing every line themselves often need time to get comfortable reviewing plans and trusting agent output rather than micromanaging it.
• Verification still matters. Artifacts make it easier to check an agent's work, but they don't replace actually reviewing what was built, especially for anything going to production.

Frequently Asked Questions

What is Google Antigravity?

Google Antigravity is an agentic software development platform that lets autonomous AI agents plan, write, test, and verify code across an editor, terminal, and browser, rather than functioning as a simple autocomplete tool.

When was Google Antigravity released?

Google announced and released the public preview of Antigravity on November 18, 2025, alongside the launch of Gemini 3. Antigravity 2.0 followed at Google I/O on May 19, 2026.

Is Google Antigravity free?

Yes, a free tier exists with access to all supported models, though rate limits are tighter than the original launch quotas. Paid tiers start at $19.99/month for Pro, with Ultra plans at $100 and $200/month for heavier usage.

Is Google Antigravity built on VS Code?

Yes, it's a fork of Visual Studio Code, though the community has debated whether the fork is direct from VS Code or via Windsurf, another VS Code-based editor. Google hasn't confirmed the exact codebase lineage.

Can Antigravity use Claude or GPT models instead of Gemini?

Yes. Alongside Gemini 3.1 Pro and Gemini 3.5 Flash, Antigravity supports Anthropic's Claude Sonnet 4.6 and Claude Opus 4.6, as well as an open-weight GPT-OSS-120B model.

What's the difference between Antigravity and Antigravity 2.0?

The original Antigravity built its Agent Manager directly into the IDE. Antigravity 2.0 is a standalone desktop application that works independently of any single editor, adding dynamic sub-agents, scheduled background tasks, and deeper integrations with Google AI Studio, Android, and Firebase.

How is Google Antigravity different from Cursor?

Cursor is editor-centric — it's built around keeping you in the loop on every suggestion. Antigravity is agent-centric, designed for delegating whole tasks to autonomous agents that work asynchronously and report back through Artifacts.

Final Thoughts

Google Antigravity represents a genuine shift in how AI-assisted coding tools are being designed — away from autocomplete-on-steroids and toward something closer to managing a team of junior developers who can plan, execute, and self-verify. Whether it's the right tool for you in 2026 depends largely on your tolerance for its still-evolving pricing model and your comfort level handing real autonomy to an agent. If you're curious, the free tier is genuinely enough to get a feel for the Manager Surface and Artifacts system before deciding whether a paid plan is worth it for your workflow.