Updated
Updated · InfoWorld · Nov 13
Google Adds Gemini 4 Hydrogen to Antigravity as Orchestrator Automates Multi-Model Coding
Updated
Updated · InfoWorld · Nov 13

Google Adds Gemini 4 Hydrogen to Antigravity as Orchestrator Automates Multi-Model Coding

3 articles · Updated · InfoWorld · Nov 13

Summary

  • Google added Orchestrator to Antigravity, letting developers run planning, coding and review in one session by handing work between Gemini planner models and the new Gemini 4 Hydrogen.
  • Hydrogen acts as the execution model—editing files, writing code, and running builds and tests—while Carbon or Argon explores the codebase, maps dependencies and revises plans when execution stalls.
  • Analysts said the setup could lift coding speed and code quality while cutting token use, giving CIOs tighter control over AI spending by reserving heavier models for reasoning-intensive work.
  • Those gains come with governance risks: Antigravity documentation does not yet detail per-step model logs, raising questions about audit trails, accountability, security controls and how each model handles source code and IP.
  • Google is offering Orchestrator with Carbon to AI Ultra subscribers and with Argon to Pro and Ultra users; Hydrogen alone has a 1 million-token context window and costs $1.50 per million input tokens and $7.50 output.

Insights

If Google's AI agents autonomously write and approve code, who takes the fall when a catastrophic vulnerability quietly slips into production?
Could Orchestrator's seamless model-switching mask a hidden cost explosion if the AI gets trapped in an endless loop of failed tests?
With per-step logs missing, how can enterprises truly trust an AI planner that hides its decision-making process in the shadows?