Google says Orchestrator, a new model option in Google Antigravity, pairs Gemini 4 Carbon with Gemini 4 Hydrogen inside a single session, and it is rolling out to Google AI Ultra subscribers. Google's X post announces the same rollout in shorter form.

How Orchestrator splits the work

Orchestrator appears in the Antigravity model dropdown. Instead of sending a whole task to one model, Carbon starts in a read-only phase to explore dependencies, resolve ambiguities and write a plan with re-planning triggers. Hydrogen then makes the edits, updates packages and runs tests. If Hydrogen hits friction, it hands back to Carbon, which inspects the state, repairs the plan and resumes. The blog contrasts this with a static model router that picks one model from the first prompt.

On Terminal-Bench 4.1, the blog says Orchestrator reached near-Carbon task success while using 24% less quota and finishing 19% faster on average. Google's X post cites Terminal-Bench 4.0 for the same figures, so this article uses the blog's 4.1.

Our earlier report on Gemini 4 Carbon's debut covers the model that does the planning.

Who can use it

  • Orchestrator: Google AI Ultra subscribers, with more models and plans promised later. Enterprises can use it in Gemini for business. Chrome Geek reports that Pro users get Hydrogen but not Orchestrator.
  • Hydrogen alone: rolling out to paid Google AI Pro and Ultra subscribers in Antigravity, according to the blog and Unite.AI.
  • Carbon: Google's Carbon post says Ultra subscribers can use it in the Antigravity App, the Antigravity CLI and IDE extensions. The Orchestrator post does not say whether Orchestrator reaches the CLI or IDE extensions.

For API use, Unite.AI reports Hydrogen at $1.50 per million input tokens and $7.50 per million output tokens. That is a separate price from the Antigravity plans.

Hydrogen on its own

When you select Hydrogen directly, you can choose among low, medium, high (the default) and a new xhigh reasoning level. The blog reports a score of 51 on the Artificial Analysis Intelligence Index, above Argon's 50, and 57% on Terminal-Bench 4.1, more than triple the 18.2% it gives for Gemini 3.8 Flash. Unite.AI notes that the model card's cheating-adjusted Terminal-Bench 4.1 table lists Claude Sonnet 5.5 at 63.9%, above Hydrogen's 57.0%.

The blog does not give the price of Google AI Ultra, and it does not say which tasks or runs produced the 24% and 19% figures.

Our read: Orchestrator's appeal is lower cost and faster runs within one session, but for now it is limited to Ultra, and the efficiency numbers are Google's own.

Sources