Google AI Studio Ultra leak points to a new Build mode

TestingCatalog reports that an unreleased Ultra option appears in Google AI Studio Build, with a description referring to advanced tools and skills. The October 9 report is a single firsthand observation. Launch timing, pricing and eligibility remain unconfirmed.

For developers, the useful question is what another mode would change in a project that already has working code. A better app builder needs to preserve existing behavior, connect the right services and leave changes that someone can inspect.

What Build already does

Google’s Build documentation describes generating web apps and native Android projects from prompts. Web apps use a React frontend by default and a Node.js server. Build already uses components of the Antigravity agent to manage project context and changes across files. Users can inspect and edit code, exchange changes with GitHub and deploy web apps to Cloud Run. These are documented capabilities of the existing product.

The full stack guide adds important detail. The agent can install npm packages, keep secrets in the server environment and configure a Firestore database and Firebase authentication. Apps can also connect to Workspace services such as Gmail and Sheets. Each person using those integrations authorizes access to their own account. Google recommends keeping interface logic separate from server logic and handling failures from external services.

That gives a future mode a concrete standard to meet. A more elaborate preview would be only one part of the result. The harder question is whether a change still works after a database write, a failed request or another round of editing.

The practical test is a working change

Consider a small inventory app with a product list, saved stock counts and a form for adding items. A useful comparison would ask each available mode to add a low stock alert while preserving the existing form and stored records. Start from the same project and give both runs the same instructions.

• Check the whole path. Adding an item should update storage and the visible list. An alert should appear at the agreed threshold and disappear when stock is replenished.
• Test a failed save. The interface should explain what happened and preserve the user’s input. Record whether the agent identifies and repairs the failure without breaking another feature.
• Review the changes. Check which files and dependencies changed, whether credentials stayed out of browser code and how much manual cleanup the project still needs.

Measure completion against those requirements. Record elapsed time, any visible usage charge and the number of corrections needed. Repeat the same task before drawing conclusions, since one attractive result cannot establish consistent behavior. This is a suggested evaluation method. ByteForward has not tested the reported mode.

Keep Argon access separate

Google announced Gemini 4 Argon on September 30. Its rollout plan begins with trusted testers and describes later access starting with paid API customers and Google AI Ultra subscribers. That announcement does not identify Argon as the model behind the reported Build option. The shared Ultra name also does not establish a subscription requirement for this mode.

What would settle the open questions

A release note should identify the available tools, access conditions and intended use. A reproducible product session should show which option was selected and what it actually changed. Until those details are available, developers can use the documented Build workflow and keep this report separate from decisions about subscriptions or model access.