Updated on September 12, 2026
Google Antigravity is Google's agentic development platform, where autonomous agents plan, write, run and test code while you supervise. It arrived in November 2025 alongside Gemini 3 and has since grown into a full family, an IDE, an agent manager app, a CLI and an SDK. Individual developers pay nothing, within weekly rate limits, and expanded access rides on Google AI subscriptions from $19.99 a month. It runs on macOS, Windows and Linux, with a personal Google account as the only entry ticket.
The Agent Manager in Google Antigravity orchestrates several agents in parallel, each in its own workspace. You keep working on the frontend, an agent refactors the API in the background, and a notification lands once its tests pass. That kind of long-running supervision is where autonomous AI agents have been heading for a while.
The other signature trick, the agent opens a real Chrome instance and actually operates the app it is building. It presses buttons, fills forms, spots the broken link and fixes it before you ever open DevTools. Every mission produces verifiable artifacts you can annotate like a Google Docs file (the agent waits for your sign-off on plans, unless you switch on the setting that hands it free rein).
The platform now ships as four products, plus official editor extensions. The IDE is a fork of VS Code, inherited from Windsurf according to several accounts, and the CLI took over from the older Gemini CLI in 2026. The name itself is a nod to the classic Python easter egg import antigravity.
Against the competition, each side keeps its strengths. Cursor stays ahead on polish and a more mature MCP ecosystem, Claude Code owns the terminal, Firebase Studio targets quick web prototyping. Google Antigravity holds its ground on parallel agents and in-browser verification, and plenty of developers keep two of these tools installed side by side.
| Piece | What it does |
|---|---|
| Antigravity IDE | Full editor forked from VS Code, agents and artifacts built in |
| Antigravity 2.0 | Desktop app for steering agents across workspaces |
| Antigravity CLI | Command-line agents, successor to Gemini CLI |
| Antigravity SDK | Hooks agents into your own systems |
| Antigravity for IDEs | Extensions for VS Code, JetBrains and Zed |
Google Antigravity is free for individual developers, with weekly rate limits on agent requests and unlimited tab completions in the editor. There is no standalone subscription, expanded access comes bundled with Google AI Pro at $19.99 a month, Ultra at $99.99 and Ultra 20x at $199.99, after the former $249.99 plan was retired in May 2026. Organizations go through Google Cloud on consumption-based billing.
Free quotas have shrunk more than once since launch, and paid credits take over when the meter runs dry, otherwise the agent halts mid-task. Google has already reshuffled this ladder at least once, so the subscription page inside your own account is the only number worth trusting before you pay.
Yes, the individual tier is free, with no credit card and no announced end date, within weekly rate limits on agent requests while tab completions stay unlimited. Google AI Pro and Ultra subscriptions raise those quotas, and extra credits can be purchased once the included allowance runs out.
Google Antigravity wins on parallel agents and automatic verification inside Chrome, while Cursor leads on polish and a more mature MCP ecosystem. Many developers simply run both, and since Antigravity's individual tier costs nothing, testing it on your own repositories settles the question for free.
The official pricing page lists Gemini 3.8 Flash, 3.7 Flash, 3.6 Flash and Gemini 3.1 Pro, alongside third-party options, Anthropic's Claude Sonnet and Opus 4.6 plus the open-weight gpt-oss-120b, all available from the free tier. What Pro and Ultra change are the weekly rate limits and a more generous pool of AI credits.
Retry the task first, then start a fresh conversation and switch models, which often works when capacity is strained. If it keeps failing, disable your MCP servers one at a time, a documented clash between Claude models and a Firebase MCP tool triggers this exact failure. A full re-authentication and a quota check handle most remaining cases.
Verdict: Supervising agents instead of typing every line is the bet Google is making here, and the free tier lets you test that bet at zero cost. Developers juggling several workstreams at once, from refactoring jobs to browser-checked frontends, will get the most out of it.
« An all-in-one creative platform that brings together more than 30 image and video models in a single interface. It includes camera controls that let you direct like a real filmmaker, characters that remain consistent from scene to scene, and ready-to-use tools for ads and short-form content »
