Antigravity 2.0 Replaces Gemini CLI on 19 May 2026: What Happens After Free Access Ends

Google pulled the rug on a tool thousands of developers had wired into their daily workflow. Antigravity 2.0 landed at I/O on 19 May 2026 as the official replacement for Gemini CLI, and the free individual-request tier that made Gemini CLI so popular closes on 18 June 2026. If you build with Google's coding agent, that date matters more than the launch hype.

Quick Answer

After 18 June 2026 the old Gemini CLI stops serving requests for free, AI Pro, and AI Ultra accounts, and you migrate to the new Go-based Antigravity CLI powered by Gemini 3.5 Flash. Flash keeps a free tier with reduced daily quotas, so casual use stays free, but heavy or always-on agent work pushes you onto a paid plan.

What actually changed

Gemini CLI was a Node-based terminal assistant. Antigravity CLI is a ground-up rebuild in Go, which makes it noticeably snappier and lets it run asynchronous, multi-agent workflows in the background. You can kick off a large refactor or a multi-step research task and keep working in the same terminal session instead of waiting on a single blocking request. It is part of the wider Antigravity platform rather than a standalone CLI, so the agent talks to a server-side harness rather than running everything locally.

The model underneath is Gemini 3.5 Flash, which also launched on 19 May 2026. Flash retains a free API tier with trimmed daily quotas, while the Pro-tier models moved to paid-only earlier in the year.

What it costs after the free tier

The Flash free tier covers light, occasional use, the kind of thing where you fire off a few prompts a day. The moment your agent runs continuously, orchestrates several tasks at once, or chews through large codebases, you hit the daily quota and need a paid plan. Worth knowing for budgeting, Gemini 3.5 Flash is priced around 1.50 US dollars per million input tokens and 9 US dollars per million output tokens on the paid API, which is a step up from the model it replaced. For an SA developer that adds up quickly on a busy day, so it is worth checking your typical token volume before committing.

What it means for SA developers

There are two sensible camps. If your usage is light and bursty, the Flash free tier may still cover you, so migrate the CLI and carry on. If you are running agents hard, the calculation shifts to whether the cloud convenience is worth a recurring rand cost paid in dollars, especially with exchange-rate drag.

The alternative many local developers are weighing is running a capable model locally on their own hardware. An AI-ready desktop with a modern NPU or a strong GPU keeps inference on your machine, with no per-token bill and no dependence on a free tier that can be retired overnight, which is exactly what just happened to Gemini CLI. You can see what that kind of machine looks like in the AI PC range at Evetech, and the most popular complete builds local devs are buying show up in the PC best sellers list.

Should you migrate now or wait?

There is no upside to waiting. The old CLI stops working for everyone on 18 June 2026, so migrating early lets you sort out config, authentication, and any broken scripts while the old tool still runs as a fallback. Test your existing workflows against the Go-based CLI now rather than discovering breakages on the cutoff day.

Frequently Asked Questions

Will my existing Gemini CLI scripts still work?

Not guaranteed. The rebuild changed enough that some setups broke on launch, so treat the migration as a real port rather than a drop-in swap. Test your scripts against Antigravity CLI before 18 June so you are not caught out.

Is Antigravity CLI free to use?

The CLI itself is the front end; the cost depends on the Gemini 3.5 Flash quota behind it. Light use fits the free tier, while heavy or continuous agent work needs a paid plan.

What model powers Antigravity 2.0?

Gemini 3.5 Flash, which launched alongside it on 19 May 2026. Flash is the fast, lower-cost tier, with the Pro models reserved for paid usage.

Can I avoid the cloud cost entirely?

Yes, by running a capable model locally on a machine with a strong GPU or NPU. That removes the per-token bill and the dependence on a free tier, at the cost of buying the hardware up front.

What happens to my data with a cloud agent?

Cloud agents send your code and prompts to Google's servers for processing. If you work on sensitive or client code, a local setup keeps everything on your own machine, which is the stronger privacy position.