Choose Gemini 4 Neon to set the thinking level yourself, or choose Orchestrator (Neon) to have Neon plan the work and Hydrogen edit code and run tests in the same session.
Suitable tasks: Feature development, refactoring, and debugging across multiple files, especially when architectural decisions can be separated from implementation.
Unsuitable tasks: For complex changes that require deep reasoning at every step, select Neon directly. For tasks that need multiple agents to work at the same time, use Antigravity's parallel execution feature.
Applicable model versions: Standalone mode uses Gemini 4 Neon. In Orchestrator mode, Gemini 4 Neon plans and Gemini 4 Hydrogen executes.
Applicable clients, agents, or APIs: The release announcement lists the Antigravity desktop app, CLI, and IDE extension. The settings below were checked against the official documentation for the desktop app and CLI.
Recommended thinking levels and settings: Standalone Neon defaults to High. Use Low for small interactive changes and Extra for deep, multistep reasoning. Orchestrator uses a fixed configuration, with no thinking level setting in the interface.
In the desktop app, open the model selector below the chat input and choose Gemini 4 Neon or Orchestrator (Neon). New conversations default to Neon if your plan includes it and you haven't previously selected a model. If you've already selected one, the app keeps your selection. Antigravity model documentation
| Mode | Model selector | CLI level |
|---|---|---|
| Standalone Neon | Low, Medium, High, Extra | low, medium, high, xhigh |
| Neon plans, Hydrogen executes | Orchestrator (Neon) | Fixed configuration |
The Extra option in the model selector corresponds to Extra High in the release announcement. In the CLI, use --model gemini-4-neon-xhigh to select that level directly. The official documentation gives the following command for starting Orchestrator in the CLI. Antigravity model documentation
agy --model "Orchestrator (Neon)"Select Orchestrator (Neon) and submit a coding task with a clear scope and a way to verify the changes. Neon first inspects the repository in read-only mode, then drafts an implementation plan and defines when to revisit it.
Once the plan is set, Hydrogen takes over to edit files, update dependencies, or run builds and tests. It checks the results against the plan. Neon can directly answer requests that only call for an explanation or review.
If verification fails twice in a row, an interface behaves differently than expected, or a design decision comes up that the plan doesn't cover, Hydrogen hands control back to Neon. Neon reviews the progress and revises the plan before Hydrogen resumes work. Antigravity model documentation
The release announcement confirms that the two models can hand off work within a single task request. The model documentation says handoffs happen automatically within a single response and the conversation uses one agent throughout. The release announcement uses /swarm in a separate game demo of the parallel execution feature. Antigravity release announcement Antigravity model documentation
The release announcement describes a gradual rollout. The model documentation lists Orchestrator (Neon) for Google AI Ultra and paid Google AI Pro. Some eligible high-end devices can also access it through a Pro trial. Enterprise plans support selecting Neon directly, and the plan table marks Orchestrator (Neon) as unavailable for those plans. Check your account's model selector for actual availability. Antigravity model documentation
We reviewed the release announcement and model documentation when collecting this note, without running the model in the Antigravity client. The source gives the configuration and handoff rules and omits the full demo input. Handoff behavior, test results, and quota usage still need to be verified in the specific account and project.
Gemini 4 Neon