Quick answer

Google Antigravity 2.0 launched at Google I/O on May 19, 2026 as a standalone desktop app where several agents plan, write, run, and verify code in parallel, with scheduled background tasks, custom sub-agent workflows, voice commands, and a Chrome extension. It ships with a Go-based CLI that fully replaces Gemini CLI (consumer access to Gemini CLI ended June 18, 2026), an SDK for embedding the agent harness, and Managed Agents in the Gemini API that run in isolated Linux environments from a single call. Gemini 3.5 Flash is the default model. A free tier signs in with a Google account; paid access comes with Google AI Pro at $20 a month and AI Ultra tiers at $100 and up. Antigravity 1.0 from November 2025 remains available.

Most "AI coding tools" are an editor or a terminal. Antigravity is the first from a major lab that is deliberately all of them at once, sharing one agent harness between a desktop app, a command line, an SDK, and an API. That is both its pitch and the source of its complexity.

The four surfaces

  • Desktop app: run multiple agents in parallel, build custom sub-agent workflows, schedule background tasks, use voice, browse with a Chrome extension
  • CLI: terminal access to the same harness, carrying over Gemini CLI's Agent Skills, Hooks, Subagents, and Extensions (now called plugins)
  • SDK: programmatic access to the harness for embedding agents in your own infrastructure, designed for Gemini models
  • Managed Agents in the Gemini API: one call starts an agent that reasons, uses tools, and runs code in an isolated Linux environment that persists across follow-up calls; custom agents extend with markdown skill files

What it costs

The free tier is real: sign in with a Google account and, per launch coverage, you get access to Gemini 3.1 Pro plus third-party models such as Claude Sonnet 4.5 and gpt-oss within limits. Google AI Pro ($20 a month) and AI Ultra ($100 a month, with a higher Ultra Premium tier reported at $200) raise the limits and add the latest models. Developers using Managed Agents through the Gemini API pay per token on the model they choose. Sources disagree on the exact top tier, so check Google's page before budgeting.

Where it is strongest

Fleets. If your work is "run six agents on six tickets and review the results", Antigravity's parallel model, scheduled tasks, and worktree-style isolation are built for it, and the Managed Agents API removes the sandbox problem that otherwise eats a week of engineering. The Gemini 3.5 Flash default is fast and cheap, and Google's claim that it beats Gemini 3.1 Pro on most benchmarks is, as always, Google's own.

Two limits to design around

  • Parallel agents are isolated: agents on the same codebase cannot see each other's file changes unless you route outputs through a shared layer, so merge conflicts are your problem
  • Managed Agent sessions expire: environments persist across follow-ups but have a time-to-live, so long-running workflows must handle session expiry explicitly
  • And a migration note: Gemini CLI users got a hard cutoff, not a gradual deprecation; scripts that called it need updating

Related reading

Muse Code vs Claude Code vs Codex CLI Best AI coding tools in late 2026 What is spec-driven development?

Bottom line

Antigravity is the most complete agent toolkit a single vendor offers, and the free tier makes it easy to try. Pick it if you are on Gemini and your work looks like orchestration. If your work looks like editing, Cursor still feels better day to day.