Google Antigravity Explained: What It Is and What It Means for Your Team

Google Antigravity is Google's new agentic development platform. Google announced it in November 2025, alongside the Gemini 3 model.

It is not just a chatbot in a sidebar. It is a full development environment built to let AI agents plan, write, and test code on their own.

For an engineering team, the practical questions are which of the four products to use, which plan fits, and what controls Gemini Enterprise adds.

What Is Google Antigravity?

Google Antigravity is an agentic development platform that includes a 2.0 desktop app, an IDE, a command-line interface, and an SDK. Google describes it as a way to "operate at a higher, task-oriented level" (Google Developers Blog).

A normal AI coding assistant suggests lines as you type. Antigravity goes further. It runs AI agents that plan a task, write the code, run it, and check the result.

Google built two ways to work inside it. The "Editor View" is a familiar AI-powered IDE with tab completions and inline commands. The "Manager Surface" is a separate screen where you spawn and watch several agents work at once (Google Developers Blog).

The agents can act across your editor, terminal, and browser. They can write code, launch the app, and test it without you stepping in at every turn (Google Developers Blog).

• Editor View: a standard AI IDE with completions and inline commands.
• Manager Surface: orchestrate and observe multiple agents working async.
• Agents act across editor, terminal, and browser.
• Cross-platform: macOS, Windows, and Linux (Google Developers Blog).

The Four Google Antigravity Products

Google Antigravity is an agentic coding platform distributed across four surfaces: a standalone desktop application, a dedicated development environment, the agy command-line interface, and a Python software development kit. Google expanded the lineup at Google I/O on May 19, 2026.

Google also released Antigravity editor extensions on August 20, 2026, per the Antigravity IDE extensions announcement. They support Visual Studio Code, Visual Studio, Zed, JetBrains 2026.2.1 and later, and Xcode 27 beta 6 and later. These editor add-ons run on every tier, including the free Individual plan.

Developers getting started with Google Antigravity can begin with the standalone desktop app, or install the agy command-line tool. You can review installation steps and permission controls in our guide on how to use Google Antigravity.

• Google Antigravity 2.0 desktop app: A standalone desktop application with no code editor. It succeeded the IDE's Agent Manager, as announced on the Google Antigravity 2.0 launch post.
• Antigravity IDE: The original coding environment, which Google recommends developers keep while Google prepares to remove the Agent Manager in an upcoming update.
• [agy command-line tool](/guides/google-antigravity-cli): A terminal tool written in Go that uses the same agent harness as the desktop app. It replaced Gemini CLI for Google AI Pro, Ultra and free individual users on June 18, 2026, per Google's Gemini CLI transition notice.
• Python SDK: An Apache 2.0 research preview installed with pip install google-antigravity. Developers use it to build custom agents with subagents, hooks, and Model Context Protocol servers, per the Google Antigravity SDK announcement.

What Are Antigravity's "Artifacts"?

Artifacts are the tangible deliverables Antigravity's agents produce as they work. Google lists task lists, implementation plans, screenshots, and browser recordings as examples (Google Developers Blog).

The idea is that you should not have to read raw logs to trust an agent. Instead you review its plan, its screenshots, and its recordings.

You can leave feedback directly on an Artifact. The agent then refines its work based on your comment (Google Developers Blog).

For a team, this matters. Artifacts create a reviewable trail of what the agent did and why. That is closer to how a human engineer documents a change.

Which AI Models Does It Use?

Google Antigravity offers Gemini, Claude, and GPT-OSS models, and which models you receive depends on your plan tier. When Antigravity launched in November 2025, the initial lineup included Gemini 3, Claude Sonnet 4.5, and GPT-OSS. The current lineup on the Antigravity models documentation spans several model families, while Nano Banana 2 handles images.

Gemini 3.8 Flash is available on every plan, while Gemini 3.7 Flash, Gemini 3.6 Flash, and Gemini 3.1 Pro are restricted to Gemini Enterprise tiers. Gemini 4 Argon rolls out on October 13, 2026, for Google AI Ultra and Enterprise subscribers in the standalone desktop app and CLI, but not in the Antigravity IDE. For Anthropic models, Claude Sonnet 5.5 and Claude Opus 5.5 require Google AI Pro (not a Pro trial), Google AI Ultra, or Gemini Enterprise.

Claude Sonnet 4.6 and Claude Opus 4.6 remain on free and Pro tiers, while GPT-OSS-120B is on free, Pro, and Ultra tiers until their scheduled removal on November 2, 2026. Google states no default model for Google Antigravity. Switching between providers inside one workflow helps you avoid vendor lock-in, as discussed in our guide to the best LLM for coding.

• Gemini 3.8 Flash: Free, Pro, Ultra, Enterprise
• Gemini 4 Argon: Ultra and Enterprise (desktop app and CLI only)
• Gemini 3.7 Flash, 3.6 Flash, and 3.1 Pro: Enterprise only
• Claude Sonnet 5.5 and Opus 5.5: Pro (non-trial), Ultra, Enterprise
• Claude Sonnet 4.6 and Opus 4.6: Free and Pro. GPT-OSS-120B: Free, Pro, and Ultra (both removed November 2, 2026)

How Much Does Google Antigravity Cost?

Google Antigravity has a free Individual plan and paid tiers: Google AI Pro at $20 per month, and Google AI Ultra at $100 or $200 per month. Organizations can buy Business seats through Google Cloud from $30 per seat per month or pay-as-you-go, and eligible Gemini Enterprise editions include Google Antigravity. The free tier includes unlimited Tab completions and Command requests with basic weekly rate limits. Google does not publish numeric token quotas or credit-pack prices, so verify current figures on the Google Antigravity pricing page.

Usage limits reset on different schedules across tiers. On Google AI Pro and Google AI Ultra, usage quotas refresh every 5 hours until you reach a weekly cap, while the free Individual tier refreshes once a week. Gemini Flash and Pro models draw from one shared quota based on API pricing, whereas non-Gemini models draw against a separate fixed limit.

Base plans do not bundle monthly AI credits, leaving extra capacity strictly to an overage setting configured to Never or Always. When set to Always, overages bill at Gemini Enterprise consumption rates without an option to bring your own API key for extra limits. For a full breakdown of team seats, budget caps, and model access, read our Google Antigravity pricing guide or the official Google Antigravity plans documentation.

Security, IP, and Data Handling: What We Can and Cannot Confirm

Google Antigravity has been included in eligible Gemini Enterprise Standard and Plus tiers since August 20, 2026. According to the Google Antigravity enterprise update, these tiers provide admin budget caps, capped overages, audit logging, and sandbox and Model Context Protocol (MCP) restrictions. Individual subscriptions on free, Pro, and Ultra tiers do not include these organization-level safeguards.

Google has not published Antigravity-specific terms covering model training on customer code, data retention schedules, or intellectual property (IP) indemnity. Ask for those terms in the Gemini Enterprise contract before regulated code goes into Antigravity.

What we can say is directional. Because Antigravity offers third-party models like Claude and GPT-OSS, your code may be sent to more than one model provider. Each provider has its own data terms.

Teams piloting Google Antigravity should run their tests through Gemini Enterprise with sandbox and MCP restrictions turned on, while keeping proprietary code out of unmanaged individual accounts. A prompt-injection flaw documented by CyberScoop allowed code execution past strict mode until Google patched the issue on February 28, 2026.

• Code-training and retention terms: not published for Antigravity specifically.
• IP indemnity: not verified for Antigravity.
• Multiple model providers may see your code, depending on the model you pick.
• Safe default: evaluate on non-sensitive code only.

How Should a Team Evaluate Antigravity?

The smartest first step is a small, time-boxed pilot on a low-risk project. Antigravity is new, so treat it as an experiment, not a standard.

Pick two or three engineers. Give them a real but non-sensitive task. Ask them to use the Manager Surface and judge whether the agents save real time.

Set a clear rule that a human reviews every agent change before it merges. Artifacts help here, but human review stays mandatory. Agentic tools can move fast in the wrong direction.

Track two things: time saved and rework caused. A tool that writes code fast but creates bugs is not a win. Compare Antigravity against tools your team already trusts.

• Time-box the pilot (for example, two weeks).
• Use non-sensitive code only until Gemini Enterprise controls are in place.
• Require human review on every agent-generated change.
• Measure time saved minus rework caused.

Who Is Google Antigravity For?

Google Antigravity fits teams that want to test agent-first development early and can plan around unpublished quota numbers. It is a strong fit if you already lean on Google Cloud and Gemini.

It is a good match for teams curious about running many agents at once, not just autocomplete. The Manager Surface is built for that style of work.

Google added administrative controls through Gemini Enterprise on August 20, 2026, but Google does not provide published intellectual property (IP) indemnity or Antigravity-specific data terms. Any team that requires written IP indemnity is a weaker fit for Google Antigravity and should request those terms directly in its Gemini Enterprise contract.

Regulated firms (finance, health, legal) should start through Gemini Enterprise, with its audit logging and sandbox and MCP restrictions, or run a sandbox-only pilot. Our AI coding assistant security and governance guide covers the controls to demand before any tool touches real code.

Verdict

Google Antigravity is one of the most ambitious agentic IDEs shipped to date. Its agent-first design, Artifacts, and model choice make it worth watching closely.

For individual developers, the free Individual plan is an easy option to test. Businesses should pilot Google Antigravity first and standardize through Gemini Enterprise for admin budget caps and audit logging, or through Google Cloud Business seats from $30 per seat per month. Google publishes no numeric quotas on its pricing page, and intellectual property (IP) indemnity terms are not detailed in these official sources.

The right move is a controlled evaluation on non-sensitive code, with human review on every change. If Google publishes numeric quotas and IP indemnity terms, revisit Antigravity as a team standard.