Free guide
How Google's Antigravity Orchestrator Works (and How to Turn It On)
Google's new Orchestrator runs two Gemini models in one coding session: a big one that plans and a fast one that does the work. Here's how the hand-off works and how to switch it on.

The short version
- Every message starts with a read-only planning phase: the planner explores your codebase and writes an implementation plan without changing anything.
- Gemini 4 Hydrogen, the fast model, then applies the edits, runs builds and tests, and checks the results against the plan.
- If a check fails twice, or something behaves differently than expected, control goes back to the planner, which fixes the plan and hands it back.
How it actually works
Think of a rally team
A rally car has two people in it. The navigator reads the route and calls out the roadbook. The driver drives fast. Orchestrator works the same way. Instead of one model doing the whole job, two Gemini 4 models share one conversation. Google's docs are clear that this isn't a router that picks one model per task: the two hand work back and forth inside a single turn.
The navigator writes the roadbook
Each message starts with a read-only planning phase. The planner, Gemini 4 Carbon (or Argon, depending on the option you pick), explores the codebase, resolves open questions and writes an implementation plan. It doesn't edit anything in this phase. For a plain question or a code review it just answers, and for a small follow-up whose steps are already settled it hands off a short plan.
The driver takes the wheel
Then Gemini 4 Hydrogen takes over. It applies the edits, writes the code, runs builds and tests, and checks the results against the plan. Hydrogen is Google's fast Gemini 4 model, so most of the turn runs on the quicker, cheaper model.
Friction sends the roadbook back
When something goes wrong, Hydrogen doesn't keep guessing. Google's docs list three kinds of friction: a verification step that fails twice, an interface that behaves differently than expected, or a design choice the plan didn't settle. Then it hands control back to the planner, which looks at where things stand, repairs the plan and hands execution back. On longer tasks, and when you send a message while Hydrogen is working, the planner also checks progress against the plan. Google says that on Terminal-Bench, Orchestrator came close to Carbon's success rate while using 24% less quota and finishing 19% faster on average.
How to set it up
1.Check your plan
Orchestrator (Carbon) needs Google AI Ultra. Orchestrator (Argon) works on Google AI Ultra and on paid Google AI Pro (or a trial on an eligible premium device). Free and Google AI Plus plans don't get either option.
2.Update Google Antigravity
Orchestrator is rolling out, so update the Antigravity app to the latest version before you look for it.
3.Pick it in the model selector
In the Antigravity app, click the model selector under the prompt box and choose Orchestrator (Carbon) or Orchestrator (Argon). It stays selected until you pick another model.
4.Or pick it in the Antigravity CLI
Run /model inside the CLI and choose an Orchestrator option (the choice sticks across sessions), or start the CLI with the model set. agy models lists the exact model names.
agy models agy --model "Orchestrator (Carbon)"5.Give it a job with a plan and a test
Orchestrator shines on work with several steps and a way to check them. Try a task like this in a repo that has tests, and watch the plan come first and the edits after.
Add input validation to the signup form, update the API handler to reject bad emails, and add tests for both. Run the tests until they pass.
The fine print
- The 24% less quota, 19% faster and near-Carbon success figures are Google's own results on Terminal-Bench, not an independent test. Google's pages name it Terminal-Bench 4.0 in one place and 4.1 in another.
- Orchestrator runs with a fixed setup: you can't choose a thinking level for it, and --effort or /effort don't apply in the CLI.
- It's still one agent. For parallel or cross-checked work, Google's docs point to subagents, Swarm, Boost or Teamwork instead.
- Enterprise access is unclear: Google's launch post says enterprises can use both options through Gemini for business, but the Antigravity models page doesn't list Orchestrator for Enterprise plans yet.
- Google says more models and plans will be supported in the future.
FAQ
What is Orchestrator in Google Antigravity?
A model option that pairs two Gemini 4 models in one conversation. A planner model writes the plan and Gemini 4 Hydrogen carries it out, handing control back to the planner when it hits friction.
Which models does Orchestrator use?
Orchestrator (Carbon) is planned by Gemini 4 Carbon and Orchestrator (Argon) by Gemini 4 Argon. In both, Gemini 4 Hydrogen does the edits, builds and tests.
Who can use Orchestrator?
Orchestrator (Carbon) is for Google AI Ultra. Orchestrator (Argon) is for Google AI Ultra and paid Google AI Pro. Free and AI Plus plans don't get it.
Does Orchestrator really save quota?
Google says that on Terminal-Bench it reached near-Carbon success rates while using 24% less quota and finishing 19% faster on average. Those are Google's own numbers, so test it on your own work.
When does Hydrogen hand the task back to the planner?
When it hits friction: a verification step that fails twice, an interface that behaves differently than expected, or a design choice the plan didn't settle. The planner repairs the plan and hands execution back.
Sources
- Gemini 4 Hydrogen and Orchestrator (Google Antigravity blog, Nov 12 2026)
- Choose a model: Orchestrator (Antigravity docs)
- Introducing Gemini 4 Hydrogen (Google blog)
- Google's launch post on X
Facts checked 2026-11-13 against Google's Antigravity blog post and models docs, the Google blog post on Gemini 4 Hydrogen, and Google's launch post on X.
Want help building with tools like this?
Join the Shipping Skool community free. Paid members get live bootcamp calls every week and weekly-updated courses.
Join Shipping SkoolMore free guides
How ChatGPT Wallet Pays Without Seeing Your Card (and How to Set It Up)
OpenAI's new ChatGPT Wallet lets your dot pay for things on your behalf, and neither the agent nor the store ever sees your real card number. Here's how the stand-in card works and how to set it up.
Claude Code Custom Themes: Animated Backdrops and Your Own Colors
The Claude Code desktop app can now wear a theme: your own colors and an animated scene behind the chat. Here is how a theme works, two ways to get one today, and why you should treat a theme like code.
Astryx: Give Your AI Agent Meta's Design System So Every Screen Matches
Meta's open-source design system ships 150+ React components, seven themes and a CLI your coding agent reads from. What it is, how it works, how to set it up today, and the fine print.