Google Antigravity 2.0: From AI-Powered IDE to Standalone Agent-First Development Platform
- May 27
- 2 min read
When Google launched Antigravity in November 2025 alongside Gemini 3, it was an AI-powered IDE with a manager surface for spawning agents. On May 19, 2026 at I/O, Google replaced that frame entirely. Antigravity 2.0 is no longer an editor. It is a standalone agent platform with four pieces, and the editor is just one consumer of it.
The Four Pieces
Antigravity 2.0 is split into a desktop app, a CLI, an SDK, and a Managed Agents service in the Gemini API. The same agent harness runs underneath all four, which is the architectural point of the rebrand.
Antigravity Desktop App: a standalone, non-IDE workspace built for orchestrating dynamic subagents, scheduled background tasks, and voice-driven workflows. Available for macOS, Windows, and Linux.
Antigravity CLI: replaces the Gemini CLI for developers who prefer terminal workflows, and preserves the previous Gemini CLI feature set while sharing the same harness as the desktop app.
Antigravity SDK: lets developers define custom agent behaviors as code and self-host them on the infrastructure of their choice.
Managed Agents in the Gemini API: a single API call spins up an isolated Linux execution environment per agent. Agents can run multi-turn sessions, use tools, and execute code in that sandbox. Custom agent definitions are authored as markdown files.
Models and Defaults
Gemini 3.5 Flash is the default model across the platform, which is consistent with Google's positioning of 3.5 Flash as fast enough to be the agent runtime workhorse. Antigravity also supports Anthropic Claude Sonnet 4.5 and OpenAI GPT-OSS, carrying forward the multi-model story from the original Antigravity.
Pricing: Welcome to AI Ultra
Google introduced a $100 per month AI Ultra tier alongside the launch. AI Ultra gives 5x the usage limits of the existing Google AI Pro plan inside Antigravity. Spark, the personal agent that ships from the same I/O keynote, will launch first to AI Ultra subscribers, which is how Google is bundling the two products.
What This Means for Agent Builders
The structural read is that Google is pulling apart agent platforms into the same layers cloud providers eventually pull every platform into: a managed service for the casual case, a CLI and SDK for the teams that want to own their own pipeline, and a desktop surface for the people whose work is the orchestration itself. The shared harness underneath is the key piece. If you are building on Antigravity, your custom subagents portably work across the CLI, the desktop app, and the Managed Agents API.
If you are building agent infrastructure outside Google, the deeper signal is in the Managed Agents shape. A single API call that returns an isolated Linux environment for tool use and code execution is the same primitive you see in Anthropic Computer Use, OpenAI Code Interpreter, and AgentCore, and it is rapidly becoming the assumed substrate that frontier agents are built on.
