Google Antigravity is Google's agentic coding platform: an AI development environment where you describe work and autonomous agents carry it out, verify it in a real browser, and hand it back for review. It ships as four surfaces, an IDE, an agent manager, a terminal CLI and a Python SDK, all sharing one quota. The free tier costs $0 and includes every product feature plus Claude Opus 4.6, which makes it the most generous free offering in this category. Above free there is no standalone price: Antigravity quota comes bundled into Google AI Pro at $19.99 a month or one of two Google AI Ultra plans at $100 and $200.

That is the short version. The longer version is more interesting, because Antigravity is not quite the product most write-ups describe, and two details that decide whether it suits a team are buried in Google's docs rather than stated on the pricing page.

What is Google Antigravity used for?

Google's own description calls it a platform for orchestrating autonomous subagents, scheduled tasks and interactive code verification. That is a better summary than "Google's answer to Cursor", even though the comparison is how most people arrive at it, because it points at what the product is actually organized around: running agents, not assisting typing.

In practice you use it for the same work you would hand any coding agent. Implement a feature across several files. Track down a failing test. Migrate a dependency. Where it differs from an autocomplete-first tool is that the unit of work is a task with a plan attached, and the agent is expected to go away and do it rather than suggest the next line.

The four surfaces, and why the distinction matters

Calling Antigravity "an IDE" is the most common mistake in coverage of it, and it leads people to compare the wrong thing. There are four pieces:

Antigravity IDE. The editor. Tab completion, natural-language Command edits, an agent side panel, and a browser the agent can drive against your running app. This is the part that competes with Cursor.
Antigravity 2.0. The Agent Manager, a command center for watching several background agents at once instead of one chat at a time. Google treats this as the primary view and the editor as one surface among several.
Antigravity CLI. The terminal client, and the documented successor path for Gemini CLI users.
Antigravity SDK. A Python SDK for programming your own agents, tools, MCP integrations and scheduled tasks. Most editor-shaped rivals have nothing equivalent.

If what you want is a better autocomplete, you will use about a quarter of what you installed. If what you want is automation you can program and schedule, the SDK and the CLI are the reason to look here rather than at the editors.

What makes Antigravity different: the browser loop

The genuinely distinctive feature gets less attention than the pricing does. In the IDE, the agent can drive a Chrome profile against your running application, take screenshots, and check that what it built actually renders and behaves. Google documents browser recordings, screenshots, and allowlist and denylist controls for what the agent is permitted to visit.

That closes a loop most coding tools leave open. The standard failure mode of an AI coding agent is not writing code that fails to compile, it is writing code that compiles, looks reasonable in the diff, and produces a broken page. An agent that opens the page and looks at it catches a class of mistake that reading the diff does not. It is not perfect verification and it will not replace your test suite, but it is a real improvement on the honor system.

Is Google Antigravity free?

Yes, and this is worth checking against anything else you have read, because Google has rewritten the free tier more than once and most articles quote a version that no longer exists. As of August 2026 the Individual plan is $0 a month with no trial clock, and it includes unlimited Tab completions, unlimited Command requests, and access to all product features including Scheduled Tasks and the CLI.

The model list on the free plan is the part that surprises people: the full Gemini line (3.7 Flash, 3.6 Flash, 3.5 Flash and 3.1 Pro) plus three third-party models, Claude Sonnet 4.6, Claude Opus 4.6 and GPT-OSS-120b. Nothing is feature-gated. The only thing the free tier limits is agent quota, which Google describes as a "meaningful quota" refreshed weekly.

What does Google Antigravity cost?

There is no Antigravity subscription. That is the structural fact people miss when they go looking for a price. Antigravity quota is bundled into Google's consumer AI plans, so what you are buying is a Google AI subscription that raises your limits:

Free: $0, weekly-refreshing quota.
Google AI Pro: $19.99 a month currently in the US, with quota refreshing every five hours until a weekly ceiling.
Google AI Ultra: $100 a month, stated by Google as 5X the Pro usage limit.
Google AI Ultra, top tier: $200 a month, stated as 20X the Pro usage limit. This was $250 until Google restructured its AI plans at I/O in May 2026.
Organization: no seat price. You need a Gemini Enterprise edition and you are billed for consumption through your Google Cloud account.

Google's I/O announcement names Antigravity explicitly in both multiplier claims, which is useful, because it means the 5X and 20X figures are about this product rather than about the Gemini chat app. Our full breakdown of Google Antigravity pricing and how it compares to Cursor puts the ladder in a table alongside the rival most people are weighing it against.

The first thing to check: Google publishes no quota number

This is the detail that should shape how you evaluate it. Nowhere in Google's documentation is there an absolute quota figure for Antigravity. The pricing page, the plans doc and the FAQ all describe limits in adjectives: "meaningful quota", "high, generous quota", "the highest, most generous quota". The only numbers that exist are the 5X and 20X Ultra multipliers, and both are measured against the Pro tier, which has no published number of its own.

So the public ladder is 1x, 5x and 20x of a quantity nobody outside Google knows. As an individual you can live with that: install it, use it for a week, see whether you hit a wall. As the person signing off on a tool for thirty engineers, it is a genuine problem, because there is no figure to put in the model and no way to answer "what happens when we double the team" except by trying it. This is the clearest single advantage Cursor still holds, since Cursor states its included usage in dollars at every tier.

The second thing to check: enterprise loses the Claude models

Read Google's model matrix column by column and there is a trap in it. Claude Sonnet 4.6, Claude Opus 4.6 and GPT-OSS-120b are available to free users, to Google AI Plus, to Pro and to Ultra. They are not available on Enterprise.

Work through what that means. A developer evaluates Antigravity on a personal free account, spends two weeks running Claude Opus through it, likes it, and recommends it. The company then buys it properly through Google Cloud, and the model that developer was actually evaluating is gone. The evaluation did not test the plan the company bought. If Claude models are load-bearing for how your team works, verify access on the exact tier you intend to purchase before you migrate anything.

Worth adding, in fairness, that the enterprise offering is strong in the ways enterprises usually care about. It runs inside your own Google Cloud project, bills through your existing Cloud billing account, supports IAM role-based access, request and response logging, and VPC Service Controls, and Google states that enterprise prompts, responses, code and telemetry are never stored outside your private environments. If your organization already lives on Google Cloud, that is a much shorter security review than most tools in this category will get.

Is Antigravity replacing Gemini CLI?

Antigravity CLI is the documented successor path, and Google has written a migration guide for exactly this. It preserves backward compatibility with the core developer-experience constructs Gemini CLI popularized, and it still consults your global GEMINI.md, so your existing instructions carry over.

What changes is small and mechanical. Local workspace skills move from .gemini/skills/ to .agents/skills/. Extensions are now called plugins. MCP servers move out of the settings file into a standalone mcp_config.json, and remote server definitions need the URI key renamed from url or httpUrl to serverUrl. Google describes the migration as largely automatic with an interactive checklist on first launch.

Should your team adopt it?

For an individual developer, the answer is close to obvious: install it, because it costs nothing and it is not a trial. There is no reason to evaluate paid tools this month without first spending a week on the best free one.

For a team, it is a real decision with real tradeoffs. In favor: the free tier lowers the cost of a pilot to zero, the Google Cloud enterprise story is genuinely mature, and the SDK opens automation that editor-only tools cannot match. Against: no per-seat price to budget from, no published quota to model against, and a model matrix that changes underneath you at the tier you would actually buy.

The thing most adoption plans underestimate is that this is a change-management problem more than a licensing one. Handing agents real work changes how code review, branching and estimation function, and teams that skip that part end up with an expensive tool and the same delivery rate. It is worth scoring how ready your process and culture actually are before you standardize a tool across engineering, because the answer often reorders the shortlist. Our guide to choosing an AI coding assistant works through the four questions that decide it, and what agentic coding means covers the workflow shift itself.

Antigravity in one paragraph

Google Antigravity is the most capable free AI coding tool available right now, and one of the harder ones to buy. Everything a solo developer needs is included at $0, including a frontier Claude model. Everything a finance team needs, a seat price and a forecastable limit, is missing. If you are one person, install it this afternoon. If you are thirty people, pilot it on free accounts, then check the enterprise model matrix and the quota question carefully before you sign anything, because those are the two places the free experience and the paid one diverge.

If what you actually want is not an editor at all but finished work arriving as reviewable pull requests, that is a different tool shape. AI coding agents that open PRs on your existing repository work alongside an editor rather than instead of one, and the two solve genuinely different problems.

Keep reading

What is Google Antigravity, and is it worth switching to?