From the source
Google Antigravity said Orchestrator, a model-dropdown option rolling out for Google AI Ultra subscribers, reached near-Carbon task success on Terminal-Bench 4.1 while averaging 24% lower quota use and 19% faster completion by switching between Gemini 4 Carbon and Gemini 4 Hydrogen within one session.
- Google says the tool targets routine feature development and debugging, where holding onto a frontier reasoning model and multiple parallel agents can slow iteration and burn frontier quota.
- A static router instead estimates difficulty from the opening prompt and commits the whole task to the cheapest model it deems sufficient, an allocation Google says suits long-running agents poorly because a task's demands change and become clear only during execution.
- Orchestrator moves in either direction: Carbon investigates dependencies, resolves ambiguities and writes a plan with conditions that trigger re-planning, Hydrogen implements edits and runs tests, and Carbon takes back control if Hydrogen meets friction.
- Developers who want explicit assignments can connect both models through the Antigravity SDK, as in Google's open-source issue-triage demonstration, where Hydrogen agents triage and code while a Carbon agent handles refactors, new features and security checks.
- Hydrogen is rolling out separately in Antigravity to paid Google AI Pro and Ultra subscribers, and enterprises can reach both models through Gemini for business; Google says more models and plans will be supported for Orchestrator later.
- Orchestrator follows the November 10, 2026 preview of Gemini 4 Carbon and /swarm, which splits tasks across named workers; Orchestrator instead changes the model used during one evolving session.