Google Antigravity: Models, Pricing, Limits & Claude 5.5

Update (9 Nov 2026): Claude Fable 5.2 and Mythos 5.2 are out - prices, access and benchmarks are in my Claude Fable 5.2 guide.

👁 127 reads

Google Antigravity is Google’s platform for programming with AI agents: a desktop app, VS Code-based editor, and command-line tool. Instead of suggesting lines of code, the agent plans the task, edits files, runs the terminal, and checks the result in a browser while you approve the plan and changes. Without a subscription it works on Gemini models with a weekly limit, while Claude Opus 5.5 and Sonnet 5.5 require paid Google AI Pro or Ultra.

What is Google Antigravity

Google unveiled Antigravity on November 18, 2025, alongside Gemini 3 Pro. The first version was a VS Code-based editor with two views: a classic editor with an agent panel and Agent Manager, where you could run multiple agents simultaneously across different projects. From the start, the agent had access to the editor, terminal, and browser.

On May 19, 2026, at Google I/O, Antigravity 2.0 arrived. Google separated out a standalone desktop app built around agent management, added Antigravity CLI (replacing Gemini CLI) and Antigravity SDK. A detailed description of this change is in the article Antigravity 2.0: from vibe coding to agent coding. Today the download page offers four products:

• Antigravity 2.0 - main desktop app: projects, parallel agents, sub-agents, background tasks;
• Antigravity IDE - VS Code-based editor with Agent panel on the right;
• Antigravity CLI - agent in the terminal, launched with the agy command;
• Antigravity SDK - programmatic access to the same agent mechanism.

Key features we verified in the documentation and changelog:

• Agent panel and agent manager - chat with the agent alongside code or manage multiple agents at once;
• plans - the /plan command makes the agent first write a plan you can read and edit before code is generated; for autonomous work use /goal;
• artifacts - implementation plans, task lists, screenshots and browser recordings that the agent delivers as work output; you can comment on them like a document;
• sub-agents - the agent divides tasks into smaller ones; since version 2.16 each sub-agent has a status card, and since 2.19.1 you can message them directly;
• browser - the agent opens the app, clicks, takes screenshots and records test sessions;
• MCP - connecting external tools via Model Context Protocol, with configurable timeout and OAuth token refresh;
• plugins, skills, rules and hooks - since 2.18.1 a plugin store in the Customization tab; rules have a separate 20,000 token budget;
• scheduled tasks and dictation - the agent can work in the background on a schedule, and you can dictate commands by voice;
• sandbox - since 2.15.1 also on Windows, since 2.16 WSL support.

How to Install and Log In

You download Antigravity from antigravity.google/download. Requirements according to Google:

1. Download the installer for your system and run it. If the system asks whether to keep both versions or replace the old one, choose replace.
2. Log in with the Google account that has your Google AI plan (or any account if you’re starting free).
3. Create a project with the folder icon in the left sidebar and add a local folder or Git repository. This defines where the agent has access.
4. Type your first command in the message field. The documentation suggests a simple Python script that fetches page text.
5. Check for updates in the Help menu, Check for Updates.

Account warning. The editor remembers the account you first logged in with. If your Google AI Pro subscription is on a different account than your personal one, Antigravity will count you against the free plan limits and won’t show Claude 5.5. In that case, log out and log in with the account that has the plan.

CLI installs with a single command (bash on macOS and Linux, PowerShell on Windows) and registers the agy command. We described the first step-by-step launch in the guide Antigravity: how to start, and the older, open-source CLI predecessor in the Gemini CLI guide.

Models in Antigravity

You choose the model from the list under the message field. In the Agent panel, the shortcut Ctrl + / opens this list. The choice applies to the entire conversation until you change it. Recently, each model has Low, Medium and High reasoning levels.

We verified this on October 3 on an account with Google AI Pro from a partner offer. Claude 4.6 has a “Notice” label, and hovering shows the message: “Sonnet 5.5 is now available on paid Pro and Ultra plans. Third-party model access will no longer be available on your current plan starting on November 2, 2026.” So on such a plan, from November there will be no Claude in Antigravity. We described the details and activation instructions in the article Claude Opus 5.5 and Sonnet 5.5 in Antigravity, and we cover the model itself in the Claude Opus 5.5 guide.

Uncertain: the plans page in the documentation describes Free and Plus plans as “Gemini models only,” while industry sites report that Claude 4.6 works in them until November 2. You can verify this most reliably in your own model selector. According to Varun Mohan from Google, Gemini 4 Argon was used internally in Antigravity, but it’s not publicly available there yet. More about it in the Gemini 4 article.

How Much Does Antigravity Cost

Antigravity has no separate subscription. Limits depend on the Google AI plan on the account you log in with. We cover pricing in the Gemini: pricing and plans article.

In the US these same plans cost $19.99 (Pro), $99.99 (Ultra 5x) and $199.99 (Ultra 20x). Google added the cheaper Ultra on May 19, 2026, and lowered the more expensive one from $249.99 at that time. We compared whether Claude access through Google AI Pro is cheaper than through Anthropic in the article Claude through Google AI Pro.

The Google AI plan also gives you Gemini in the app, where since October 9 free accounts only have Flash-Lite. We described this in the article Gemini free from October 9. This change affects the Gemini app, not Antigravity.

Limits

Google doesn’t publish limits in number of messages or tokens, so we don’t provide such numbers. Here’s how the limits are structured:

• no plan and AI Plus - limit renewed once a week;
• Pro - limit renewed every 5 hours, until the weekly limit is exhausted;
• Ultra - highest limit, renewed every 5 hours, and highest weekly limit;
• Gemini - Pro and Flash have one shared pool, consumed according to API pricing, so the more expensive model uses it faster;
• third-party models - Claude and GPT-OSS have a separate, fixed limit because Google has less compute for them; exhausting Claude doesn’t block Gemini;
• AI credits - since May they’re not part of plans; in Pro and Ultra they work as a top-up beyond the limit, and in the AI Credit Overages setting you choose whether to use them automatically or wait for renewal;
• code completion (Tab) - unlimited in all plans.

On forums and in reviews (including XDA), users complain that the free limit runs out quickly with serious work, and in Pro there are occasional multi-day blocks after exhausting the weekly limit. These are reports, not official data. Unused limits don’t carry over to the next period. Everything known about limits and how to conserve them is described in the article Antigravity limits - how much really.

How We Work with the Agent

What saves us the most time isn’t the agent’s speed, but seeing its plan before it touches files. Our workflow:

1. Project and scope. We add only the folder the agent should work on to the project. The rest of the disk stays outside its reach.
2. Model. Ctrl + / and selection: Gemini Flash for minor fixes, Gemini 3.1 Pro or Claude Sonnet 5.5 for regular work, Opus 5.5 for difficult bugs or changes across many files.
3. Plan. We start with /plan and a task description: what should work, what not to touch, how to verify the result.
4. Plan review. We read the plan as an artifact, correct steps and add comments. The agent incorporates them without re-explaining the entire task.
5. Execution. For larger tasks the agent launches sub-agents. Their cards show what’s running, what’s waiting, and what’s finished.
6. Verification. We have the agent run tests and check the app in the browser. Screenshots and session recordings remain as artifacts.
7. Change review. We go through the file differences and only then accept. If the conversation went in the wrong direction, since 2.19.1 you can revert just the conversation without reverting files.

The /goal command lets you hand the agent an entire task for independent execution. We only use it in projects with version control, because autonomous work makes unwanted changes more likely.

Antigravity vs Cursor vs Claude Code vs Codex

We list competitor prices in USD, as we don’t have local price lists for them. They come from September 2026 comparisons and may change.

GitHub Copilot is a cheaper entry option (Pro $10), convenient for people staying in VS Code or JetBrains. Detailed competitor descriptions: Cursor, Claude Code and Codex. A comparison of all tools is in the AI for programming hub.

Who is Antigravity For

• For people already paying for Google AI Pro - at the same price they get Gemini, Claude 5.5 in the editor and disk space.
• For those who want to start free - the free plan with Gemini is enough for learning and small projects.
• For front-end and web apps - the agent checks results in the browser itself and delivers screenshots.
• For working on multiple tasks at once - the 2.0 app is built around parallel agents.

It makes less sense for people who work mainly in the terminal on Claude models (Claude Code gives more control) or need predictable limits in numbers, because Google doesn’t publish them.

Frequently Asked Questions

Antigravity - what is it?

It’s Google’s platform for programming with AI agents: the Antigravity 2.0 desktop app, Antigravity IDE based on VS Code, CLI and SDK. The agent plans tasks, writes code, uses the terminal and browser, while you approve the plan and changes.

Is Antigravity free?

Yes, you can use it without a subscription. The free plan gives you Gemini models with a weekly-renewed limit and unlimited code completion. Claude 5.5 and higher limits require paid Google AI Pro or Ultra.

How much does Antigravity cost?

Antigravity has no separate price. Higher limits come with a Google AI plan: Pro for $19.99 monthly, Ultra for $99.99 (5x Pro limit) or $199.99 (20x Pro limit). Google AI Plus has the same Antigravity limit as the free plan.

Does Antigravity have Claude?

Yes. Claude Opus 5.5 and Sonnet 5.5 work in paid Google AI Pro and Ultra, without an Anthropic API key. Trial periods don’t include them. Claude Opus 4.6 and Sonnet 4.6 disappear on November 2, 2026.

What are the limits in Antigravity?

Without a plan and in AI Plus the limit renews weekly, in Pro and Ultra every 5 hours up to the weekly limit. Gemini has a shared pool, Claude a separate one. Google doesn’t publish limits in number of messages.

Does Antigravity work on Linux and older Windows?

It works on Linux with glibc 2.28 or newer (e.g. Ubuntu 20+) and on Windows 10 and 11 in 64-bit, also on ARM. On macOS you need system 12 Monterey or newer.

Why don’t I see Claude 5.5 despite having Pro?

Most often Antigravity is logged in with a different Google account than the one with the subscription. The second reason is a trial period - then Claude 4.6 shows a “Notice” label. Finally, update the app: Help, Check for Updates.

Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →

Sources and Verification Date

• Antigravity: changelog (version 2.19.1 from 09/30/2026)
• Antigravity: model list
• Antigravity: plans and limits
• Antigravity: plan changes from 05/19/2026
• Antigravity: download and requirements
• Antigravity: getting started
• Google: developer highlights from I/O 2026
• Google: Gemini 3 and Antigravity launch
• Startup Fortune: Claude Opus 5.5 and Sonnet 5.5 in Antigravity
• NPowerUser: Claude 5.5 in Antigravity and Gemini changes
• XDA: Antigravity test and limits

We verified this on October 3, 2026 in the documentation and in our own Antigravity. We update the article after every plan and model change.

Kinetyra