Google Antigravity 2.0 Launches at I/O 2026 — Gemini CLI Dies June 18, Agent IDE War Has a New Entrant
Google used I/O 2026 to retire Gemini CLI and ship its replacement: Antigravity 2.0, a full agentic development platform that competes directly with Anthropic’s Claude Code and OpenAI’s Codex. The migration deadline is June 18 — Gemini CLI stops working that day.
What Antigravity 2.0 Is
Antigravity 2.0 is a VS Code fork rebuilt around agents. The interface has two surfaces that Google treats as distinct:
Editor View: A standard AI-powered IDE with tab completions and inline commands. Familiar to anyone who’s used VS Code or Cursor.
Agent Manager: The new surface. It’s a dedicated interface for spawning, monitoring, and directing multiple agents running asynchronously across different workspaces. Google describes it as “mission control.” Agents can work in the background — reproducing an issue, writing tests, implementing a fix — while the developer stays on other tasks.
Agents communicate via Artifacts: concrete deliverables like screenshots, implementation plans, and browser recordings. Developers review Artifacts and leave feedback directly on them, without interrupting execution.
The CLI and SDK
Alongside the desktop app, Google shipped:
- Antigravity CLI — terminal access to the same agent harness, for developers who prefer command-line workflows. Shares the same underlying model layer as the desktop, so improvements propagate to both.
- Antigravity SDK — programmatic access to Google’s agent harness for custom workflows and integrations.
- Managed Agents in Gemini API — hosted execution without local installation, available via the Interactions API and Google AI Studio.
All managed agents run on Gemini 3.5 Flash, which launched the same day.
Pricing
Antigravity 2.0 is free for individuals in public preview. A $100/month tier ships alongside it with higher usage limits. There is no published enterprise pricing or formal governance documentation yet — a gap that procurement teams at large companies will notice.
Model Optionality
Unlike Claude Code (Anthropic models only) and Codex (OpenAI only), Antigravity ships with model optionality: Gemini 3 Pro, Claude Sonnet 4.5, and GPT-OSS are all supported. Whether this is a durable commitment or a launch positioning move isn’t clear from the I/O announcements.
Security
Antigravity’s permission model mirrors Claude Code’s: a configurable allow/deny list for terminal command execution, plus a Browser URL Allowlist for the built-in browser agent. The default setting can be toggled between “always proceed” and “request review before each terminal command.” Browser agents get their own specialized model optimized for operating on live pages.
The Competitive Picture
The agent IDE market as of May 20 has four serious players:
| Platform | Model | Price (individual) | Multi-agent |
|---|---|---|---|
| Claude Code | Anthropic only | Consumption-based | No |
| Codex | OpenAI only | $100/mo (Pro 5X) | No |
| Cursor 3 | Multiple | $20/mo | No |
| Antigravity 2.0 | Multiple | Free (preview) | Yes |
Google’s multi-agent Manager surface is the clearest differentiation at launch. None of the other platforms ship a dedicated orchestration layer for parallel agents. Whether the execution matches the interface is a separate question — Antigravity 2.0 is in public preview, not GA.
What Developers Need to Do Now
Gemini CLI stops working June 18, 2026. Migration path:
- Install Antigravity CLI (
agycommand after installation) - Or install the desktop app
- Existing Gemini CLI configurations are not automatically migrated — projects using it programmatically need to update to the SDK
The transition has a 29-day window from the I/O announcement date.