Google Antigravity is a new AI-powered coding agent that flips the traditional autocomplete model on its head, putting an artificial intelligence agent in control while developers review its output. The platform, which entered public preview on November 18, 2025, is built on Google's Gemini 3 model but also supports Claude Sonnet 4.5 and OpenAI's GPT-OSS, giving developers flexibility to switch between engines mid-project.

What Makes Google Antigravity Different From Other AI Coding Tools?

Unlike traditional AI-assisted editors that function as autocomplete plugins bolted onto your existing workflow, Antigravity puts an agent in the driver's seat and asks you to review its output rather than type every line yourself. The platform runs as a standalone desktop application on macOS, Windows, and Linux, not as an extension.

The interface is built around two distinct surfaces. The Editor View looks and feels like the code editor you already know, while the Manager Surface is built for spawning, orchestrating, and watching multiple agents work at once, often on different parts of the same codebase. Google also gives agents control of a real browser, so they can open your running app, click through it, and check their own work instead of just claiming a task is done.

How to Get Started With Google Antigravity in 12 Steps?

• Download and Install: Start at the official Antigravity download page, which detects your operating system and offers the matching build for macOS, Windows, or Linux. Installation follows the pattern of any modern desktop IDE, with the first launch taking longer than a typical editor because Antigravity needs to authenticate and initialize both the Editor View and Manager Surface.
• Sign In and Understand Your Access Tier: Open Antigravity and sign in with a Google account. During public preview, that sign-in alone unlocks Gemini 3 Pro, Claude Sonnet 4.5, and GPT-OSS at no charge, subject to capacity-based rate limits that refresh every five hours.
• Recreate Your Editor Setup: Because Antigravity sits on a modified Visual Studio Code foundation, most of your muscle memory transfers directly. Panel layout, the command palette pattern, and general navigation should feel close to home from the first session. Budget a few minutes early on to reinstall your regular extensions by hand and reapply your preferred color theme.
• Tour the Interface: Understanding the split between Editor View and Manager Surface early saves confusion later. The Editor View handles traditional code editing, while the Manager Surface lets you spawn and orchestrate multiple agents working in parallel.
• Run Agents in Parallel: One of Antigravity's key features is the ability to let multiple agents work on different parts of the same codebase simultaneously, with the Manager Surface providing visibility into all running tasks.
• Let Agents Test Their Own Code: Agents can open your running app in a browser, click through it, and verify their own work, reducing the need for manual testing of agent-generated changes.

What Are the Pricing Tiers and Usage Limits?

Google offers three access tiers for Antigravity. The public preview tier is free and gives access to Gemini 3 Pro, Claude Sonnet 4.5, and GPT-OSS, gated by capacity, with rate limits that refresh every five hours. Google AI Pro costs $19.99 per month and provides the baseline Antigravity usage limit. Google AI Ultra costs $249.99 per month and provides five times the usage limit inside Antigravity compared with AI Pro, according to Google's developer messaging.

For a first project, the free preview tier is enough to complete every step in the tutorial, including building a working project from a single prompt. Developers should upgrade only once they find themselves hitting the five-hour rate limit regularly, which tends to happen when running several agents in parallel on a large repository rather than during a single small task.

What Hardware and Prerequisites Do You Need?

Google has not published strict minimum hardware specs for Antigravity during its public preview. Since the heavy lifting happens in Google's cloud rather than on your laptop, the local requirements are closer to running any modern editor than to running a local large language model (LLM), a type of artificial intelligence trained on vast amounts of text data. Here is what you need before getting started:

• Operating System: macOS, Windows, or Linux are all supported builds on the official download page.
• Google Account: Required to sign in and claim public-preview access to Antigravity.
• Internet Connection: Required at all times, since model inference runs in the cloud and Antigravity is not usable fully offline.
• Disk Space: A few hundred megabytes for the application itself, plus normal project storage for whatever repository you open.
• Git Installation: Recommended so you can diff, stage, and roll back agent-generated changes like any other commit.
• Editor Experience: Not required, but familiarity with Visual Studio Code shortcuts and panel layout carries over directly.

Google's messaging is explicit that the free preview tier is a snapshot rather than a permanent offering. Developers should treat current pricing as subject to change once Antigravity exits public preview.

How Does Model Optionality Work in Antigravity?

One of Antigravity's key features is what Google calls model optionality. Developers are not locked into Gemini 3 Pro. Claude Sonnet 4.5 and OpenAI's GPT-OSS are both available in the same interface, so you can switch engines mid-project if one is rate-limited or better suited to the task in front of you. This flexibility allows developers to optimize for different types of coding tasks without leaving the application.

The platform's capacity-gated access during public preview means usage is granted "to the degree we have capacity," according to Google's launch post. This is worth planning around before you kick off a long agent run, particularly if you are working on a large project or running multiple agents in parallel.