Summary
Google announced Gemini 4 Hydrogen on November 12, 2026, positioning it for coding and agentic developer workflows. The model has four configurable thinking levels and is available through several Google developer products.
Google announced Gemini 4 Hydrogen on November 12, 2026, saying developers can start using the model for coding and agentic workflows. The company describes it as a model for rapid iteration, including work on code repositories, debugging, interactive web development and multimodal tasks.
Google says Hydrogen can be used on its own at four configurable thinking levels or alongside its Gemini 4 Carbon and Gemini 4 Argon models. It presents the model as following instructions more closely and producing fewer fabricated claims, and says this can help developers produce cleaner, production-ready code faster.
Workflows and model options
The announced uses range from repository coding and debugging to interactive web development. Agentic workflows generally involve a model contributing to tasks that require multiple steps, rather than responding only to a single prompt. The announcement does not detail the specific actions Hydrogen can take within such workflows.
The four thinking levels give developers a way to configure the model’s approach, though Google’s announcement does not describe the settings or how they affect speed and output. Google says Hydrogen can also work alongside its larger Gemini 4 models; it points to an Orchestrator feature in Google Antigravity that pairs Hydrogen with Gemini 4 Carbon.
Where developers can access Hydrogen
Google lists the Gemini API in Google AI Studio and Android Studio, Gemini for business, and Google Antigravity as access routes. In Antigravity, the announcement says Hydrogen is available to paid Google AI Pro or Ultra subscribers. Google directs developers to its Gemini API “What’s New” guide for model details, setup instructions and controls for the thinking levels.
The launch post presents improved instruction following, fewer fabricated claims and faster production-ready coding as Google’s performance claims; it provides no benchmark measurements for them.