Google adds multi-model Orchestrator to Antigravity for coding

Coordinating AI for complex code

Google is introducing a new feature called Orchestrator to its Antigravity development environment. This tool is designed to help enterprises improve software development efficiency by coordinating multiple Gemini models simultaneously. It allows developers to delegate different parts of a project, such as planning, coding, and reviewing, to different specialized models.

The system works by combining a planning model with a lighter execution model. For example, a developer can use Gemini 4 Carbon or Argon to explore a codebase and identify dependencies. Once a plan is established, the system hands off the work to Gemini 4 Hydrogen to perform the actual coding and testing.

This approach differs from standard model routing, which usually picks one model for an entire prompt. Orchestrator instead transfers control back and forth as the work progresses. If the execution model hits a snag, like a repeated test failure, it hands control back to the planner to revise the strategy.

Productivity and cost efficiency

The primary goal of this multi-model approach is to boost developer productivity on complex requirements. Analysts suggest that this setup acts like pairing a senior architect with a fast coder. By separating high-level reasoning from the manual labor of writing code, the system can improve debugging and code quality.

Beyond speed, the feature is intended to help companies manage their AI spending. Using a single, high-capability model for every minor task can be expensive and inefficient. Orchestrator helps control costs by automatically assigning routine tasks to lighter models while reserving powerful models for difficult logic.

For CIOs managing large engineering teams, this provides a way to scale AI-assisted development. It helps prevent teams from quickly exhausting usage limits or facing unpredictable costs. By optimizing token consumption, the tool makes it easier for large organizations to integrate AI into their standard workflows.

Governance and security hurdles

While the tool offers efficiency, it also introduces potential challenges for enterprise security and auditing. Coordinating multiple models can make it difficult to track which specific AI made a particular decision. Without clear logs, it may be hard for companies to investigate errors or establish accountability for generated code.

Security is another consideration because each model brings its own behaviors and risks when accessing source code or production credentials. Experts suggest that enterprises should provide coding agents with dedicated identities and restricted credentials. They also recommend requiring human approval before any AI-generated code changes are merged into a main system.

Finally, companies need to verify how each model handles intellectual property. This includes checking where prompts and source code are processed and what data is retained. Organizations should also ensure their enterprise agreements cover every specific model that the Orchestrator is capable of invoking during a session.

Availability and pricing

Developers can currently access Orchestrator through a model selector located below the conversation prompt box in Antigravity. The specific configuration remains active throughout a conversation until the user chooses a different model. Google currently offers different combinations based on the user's subscription level.

Google AI Ultra subscribers can use Orchestrator with Carbon as the planner model. Meanwhile, both Google AI Pro and Ultra subscribers can use the configuration that features Argon as the planner. Developers also have the option to use Gemini 4 Hydrogen as a standalone model for tasks that do not require a planner.

The pricing for Gemini 4 Hydrogen is set at $1.50 per million input tokens and $7.50 per million output tokens. This pricing structure is part of the broader effort to provide predictable costs for high-volume coding tasks. These options allow teams to choose the best balance of performance and price for their specific needs.