Google Antigravity 2.0: Agentic Coding Platform 2026

TL;DR

Google Antigravity 2.0, announced at Google I/O 2026 on May 19, turns Antigravity from an agent-first IDE into a full development platform. It now spans a new standalone desktop app, a terminal tool (the Antigravity CLI, invoked as agy), an SDK, and a Managed Agents tier inside the Gemini API.¹ All of it runs on Gemini 3.5 Flash, Google's new fast frontier model. Google is also retiring the standalone Gemini CLI — it stops serving most users on June 18, 2026 — and pushing developers to migrate.² A new $100/month Google AI Ultra plan slots in below the top tier, which dropped from $250 to $200.³ The headline shift: Antigravity no longer centers a code editor. It centers the agent.

What You'll Learn

• What Google Antigravity 2.0 is and how it differs from the original 2025 release
• The Antigravity surfaces: desktop app, CLI, SDK, and enterprise platform
• Why Google is retiring the Gemini CLI — and the June 18, 2026 deadline
• How Managed Agents work in the Gemini API
• Antigravity 2.0 pricing and the reshuffled Google AI subscription tiers
• How Antigravity 2.0 compares to Claude Code and Cursor

What Is Google Antigravity 2.0?

Google Antigravity is Google's agent-first development platform — a tool for taking an idea and turning it into a production-ready app by handing the work to autonomous AI agents.¹ Antigravity 2.0, announced at Google I/O 2026 on May 19, is the platform's biggest change since it launched on November 18, 2025.¹⁴

The original Antigravity shipped as a heavily modified fork of Visual Studio Code: an AI-powered editor with an agent manager attached, released in a free public preview and pitched as Google's answer to Cursor.⁴⁵ It let you delegate coding work to agents that planned, executed, and verified tasks across the editor, terminal, and browser, producing "Artifacts" — task lists, implementation plans, screenshots, and browser recordings — so you could audit what each agent actually did.⁵

Antigravity 2.0 reframes the whole product. Rather than ship an updated editor, Google released what it calls "a new standalone desktop application" that "fully delivers on an agent-optimized experience" — a central home for orchestrating multiple agents running tasks in parallel, rather than an editor you happen to chat with.¹ The original VS Code-based Antigravity IDE still exists as a separate download with no forced sunset, but the 2.0 app is the agent-first surface Google is steering developers toward. It adds dynamic subagents for parallelized workflows, scheduled tasks for background automation, native voice commands, and built-in integrations with Google AI Studio, Android, and Firebase.¹⁴

The bigger change is scope. Antigravity is no longer a single app — it is a platform with several surfaces, all sharing one agent harness.

From IDE to Platform: Desktop App, CLI, and SDK

Under the banner of "the Google Antigravity Ecosystem," Google laid out four ways to work with Antigravity, plus a separate API path covered in the next section.¹

The Antigravity 2.0 desktop app is the new standalone application. It acts as the control surface for agents: you describe what you want built, and multiple agents — including dynamic subagents — execute in parallel while you review their progress and Artifacts. Scheduled tasks let agents run in the background, and the app ties into AI Studio, Android, and Firebase so a project can move between prototyping and production without leaving the tool.¹

The Antigravity CLI is a new terminal tool, invoked as agy, launched as part of the Antigravity 2.0 platform. It is a lightweight, high-velocity surface for developers who would rather stay in the shell — you can create new agents instantly with no graphical interface.¹ It is also the centerpiece of a deprecation Google announced the same week, covered below.

The Antigravity SDK gives you programmatic access to the same agent harness that powers Google's own products. Optimized for Gemini models, the SDK lets you define custom agent behaviors and host them on infrastructure of your choosing — useful if you want Antigravity-style agents inside your own application rather than inside Google's app.¹

Finally, Antigravity in the Gemini Enterprise Agent Platform lets Google Cloud customers connect Antigravity directly to their Google Cloud projects, aimed at teams that need agents wired into existing enterprise workloads.¹

The common thread is the agent harness. The same technology and infrastructure runs across all four surfaces — and beyond them, since Google's new Gemini Spark assistant runs on the same harness too. It is co-optimized with Gemini 3.5 Flash, the fast frontier model Google launched at the same event, which Google says was itself co-developed using Antigravity.¹⁴ Google claims 3.5 Flash outperforms its larger Gemini 3.1 Pro across almost all benchmarks while running about four times faster than other frontier models — the speed Google wants for hours-long, multi-step agent runs.¹ It is not a budget engine, though: on the Gemini API, 3.5 Flash runs $1.50 per million input tokens and $9 per million output tokens.⁶

Antigravity CLI: Google Is Retiring Gemini CLI

The Antigravity CLI does not arrive alone — it arrives as a replacement. In a separate announcement, Google said it is transitioning the standalone Gemini CLI to the Antigravity CLI and encouraging all Gemini CLI users to migrate.¹²

The deadline is concrete. On June 18, 2026, Gemini CLI and the Gemini Code Assist IDE extensions will stop serving requests for Google AI Pro and AI Ultra subscribers, as well as for individuals using Gemini Code Assist for free.² Organizations on a Gemini Code Assist Standard or Enterprise license, or using Gemini Code Assist for GitHub through Google Cloud, keep their access unchanged — the retirement targets individual developers, not enterprise contracts.²

Google's stated reasoning is consolidation. The company says developers increasingly need multiple agents that communicate with each other to split up work, which means terminal tools should share one backend with the rest of the workflow. Rather than maintain Gemini CLI and Antigravity as separate products, Google is "unifying our efforts" into Antigravity and its server-side agent harness.²

Migration will not be seamless. Google acknowledges there is no 1:1 feature parity "right out of the gate," though it says the Antigravity CLI keeps the features Gemini CLI users relied on most: Agent Skills, Hooks, Subagents, and Extensions — the last now rebranded as Antigravity plugins.² Coverage of the change has been blunt about what it is: trade press described it as Google nudging — or pushing — developers off one tool and onto another.⁷ If you script against Gemini CLI today, the migration is now on your roadmap whether you wanted it there or not.

Managed Agents Come to the Gemini API

The fifth path to Antigravity's agent harness is not an app at all. Google introduced Managed Agents in the Gemini API — a way to run an Antigravity-style agent with a single API call.¹

With one call, you can spin up an agent that reasons, calls tools, and executes code inside an isolated Linux environment. The agent runs on the same Antigravity agent harness, built on Gemini 3.5 Flash, and is available through the new Interactions API and in Google AI Studio.¹

Two details matter for developers. First, the environments are persistent and isolated: each interaction creates an environment you can resume in later calls with all files and state intact, which makes multi-turn sessions practical without rebuilding context every time. Second, you can define your own agents — extend the base Antigravity agent with custom instructions and skills written as plain Markdown files, and start from custom agent templates in the AI Studio Playground.¹

Managed Agents put Google in the same territory as other labs shipping hosted, multi-step agent runtimes — a market where Anthropic has been building out its own managed multi-agent orchestration. The pitch is identical across vendors: skip the infrastructure work of sandboxing, state management, and tool wiring, and rent the harness instead.

Antigravity 2.0 Pricing and the New AI Ultra Tier

Google reshuffled its consumer AI subscriptions at I/O 2026, and the changes directly affect how much Antigravity you get.³

There are now four paid tiers:

⚠ Prices change frequently. The values above are for illustration only and may be out of date. Always verify current pricing directly with the provider before making cost decisions: Anthropic · OpenAI · Google Gemini · Google Vertex AI · AWS Bedrock · Azure OpenAI · Mistral · Cohere · Together AI · DeepSeek · Groq · Fireworks AI · Perplexity · xAI · Cursor · GitHub Copilot · Windsurf.

The new $99.99 tier — which Google markets as "$100" — is the notable addition. Google positions it for developers, technical leads, and advanced creators, and it bundles a 5× higher Antigravity usage limit than AI Pro, 20 TB of storage, and access to Gemini 3.5 Flash.³ At the same time, the existing top-tier AI Ultra plan dropped from $250 to $200 a month while keeping its 20× usage limit.⁴³

Google also dangled a launch incentive: new and existing AI Ultra subscribers were offered $100 in bonus Antigravity credits that kick in after a plan's quota is exhausted — an offer Google said expires May 25, 2026.¹

Antigravity still has a free tier, and tab completion and basic editor features remain available without a subscription. But it is worth knowing the history before you rely on free usage: Google has trimmed Antigravity's free-tier agent limits more than once since the November 2025 launch, and introduced a credits system in early 2026. The free tier is fine for evaluation; it is not a stable foundation for daily agent-heavy work, which is precisely the gap the new $100 plan is designed to fill.

Antigravity 2.0 vs Claude Code vs Cursor

Antigravity 2.0 competes in a crowded field. The two reference points most developers will reach for are Anthropic's Claude Code and Cursor — and the three now look strikingly similar on price while differing on shape.

The pricing convergence is real: all three top out at $200/month for individuals, and all three start their paid plans at roughly $20.³ Where they diverge is philosophy.

Cursor is still an editor first. If you live in VS Code, your extensions, keybindings, and muscle memory carry over, and the agent is a powerful layer on top of a familiar surface. Claude Code is the opposite — terminal-native, GUI-free, and strongest for power users automating large codebases through scripts, CI, and scheduled jobs. Antigravity 2.0 plants itself between them by refusing to pick: it offers a desktop app for people who want a visual control room, a CLI for terminal users, and an SDK for teams who want to embed the agents elsewhere.

That breadth is the bet. The risk is that a tool trying to be all three surfaces has a steeper learning curve than a tool that does one well — early hands-on reviews already note that Antigravity asks more of new users than Cursor does. The honest answer to "which is best" is unchanged from the rest of this category: it depends on whether you think in editors, terminals, or APIs.

The Windsurf Connection

There is a backstory worth knowing. Antigravity 2.0 was presented at I/O by Varun Mohan, a Director of Software Engineering at Google DeepMind — and the former co-founder and CEO of Windsurf, the AI coding startup formerly known as Codeium.¹⁸

In July 2025, Google struck a $2.4 billion deal to bring Mohan, co-founder Douglas Chen, and roughly 40 members of Windsurf's research team into Google DeepMind, along with a nonexclusive license to certain Windsurf technology.⁸ That deal came together only after a reported $3 billion acquisition of Windsurf by OpenAI fell apart.⁸ Less than a year later, the team that built Windsurf is running Google's Antigravity platform — and the family resemblance shows in how Antigravity is built around managing fleets of agents.

It is a useful reminder that the agentic coding race is not only about models. The talent that knows how to wrap those models in a usable agent harness is scarce, and Google paid handsomely to bring it in-house.

Should You Switch to Antigravity 2.0?

For most developers, the move is not urgent — but two groups should act now.

If you use the Gemini CLI, the June 18, 2026 cutoff is a real deadline. Plan the migration to Antigravity CLI, test that Agent Skills, Hooks, and Subagents behave as your scripts expect, and budget time for the missing parity Google has acknowledged.²

If you are building agents into your own product, the Managed Agents API and the Antigravity SDK are genuinely new options. Renting a hosted, sandboxed agent harness with persistent environments is worth evaluating against the cost of building your own.

For everyone else — developers happy in Cursor, Claude Code, or another tool — Antigravity 2.0 is worth a look, not an immediate switch. Use the free tier to run a real task, not a demo, and measure whether multi-agent parallelism actually saves you time on your codebase. Antigravity's demos show agents building whole apps from a single prompt; your Tuesday-afternoon bug fix is the test that matters.

The Bottom Line

Antigravity 2.0 is a repositioning as much as a release. Google took an agentic IDE that competed feature-for-feature with Cursor and rebuilt it as a platform play: one agent harness exposed through a standalone desktop app, a CLI, an SDK, and an API. The retirement of the Gemini CLI makes the strategy explicit — Google would rather maintain one agentic surface than several.

The pieces are coherent, and Gemini 3.5 Flash gives the platform a fast, capable engine. But the proof is in the work, not the keynote. If you are on the Gemini CLI, mark June 18, 2026 on your calendar. If you are building agents, evaluate the SDK and Managed Agents against rolling your own. And if you are just curious, run one real task through the free tier and judge it on your own numbers — that has always been the right way to choose a coding tool, and a platform launch does not change it.