Google Launches Gemini 4 Hydrogen

Google released Gemini 4 Hydrogen on Wednesday, designating it as the default engine for AI Mode. The company stated that the model is engineered to parse complex inquiries that contain multiple distinct parts, obscure typos, or highly specific parameters. By breaking these requests into smaller components, the system addresses each segment within a single response. This capability is now active in search, the Gemini app, and all tiers of Google AI Studio.

Google Search product vice president Robby Stein characterized the update as the largest model upgrade to Search yet. He noted that the system is being deployed for the most difficult questions in AI Overviews. The company claims the new architecture conducts simultaneous queries across relevant subtopics, which results in a wider variety of source links. Responses also feature more visual elements, including images, comparison tables, graphs, and maps.

Stein pointed to a specific demonstration where the AI generated an interactive animated driving simulation. This response addressed a user query regarding how smartphones identify highway lanes in areas without cell service. Google also reported improved language comprehension across all supported AI Mode languages. The model replaces Gemini Flash 3.8 as the primary comparison point for performance metrics.

The release introduces four thinking level options for users: low, medium, high, and a new xhigh mode. Google asserts that the new model delivers major improvements in both speed and intelligence compared to its predecessor. The company positioned Gemini 4 Hydrogen against competing models in its class, specifically naming Claude Sonnet 5.5 and GPT-6.1 Terra as benchmarks. The new system is available in the Gemini API, Gemini for business, and Google Antigravity for AI Pro and Ultra subscribers.

The rollout marks a shift in how Google handles multi-part prompts in its search infrastructure. By prioritizing simultaneous subtopic queries, the system aims to reduce the need for users to refine their initial search terms. The addition of the xhigh mode provides a higher processing ceiling than previous versions offered. These features are integrated into the existing search interface without requiring new hardware or software installations for the general public.