Google Rolls Out Gemini 4 Hydrogen Model to Power Search Upgrades

OPA WIRE, Nov 12 (OPA) — Google announced the release of its Gemini 4 Hydrogen model, describing the update as its most significant modification to Search to date. The model is designed for rapid execution across agentic workflows, completing tasks 2.5 times faster than its predecessor, Gemini 3.8 Flash. Despite its speed, Hydrogen outperforms the more powerful Gemini 4 Argon on the Artificial Analysis Intelligence Index benchmark at its default high thinking level.

In Search, Hydrogen improves intent recognition, allowing users to input natural train-of-thought queries with multiple parts or specific parameters. The model also navigates typos more effectively and enhances Search's fan-out technique, which divides single prompts into multiple subqueries to surface overlooked websites. Furthermore, the update introduces more images, structured responses, comparison tables, and custom tools to AI Mode.

Hydrogen operates independently on programming and testing tasks, or works alongside frontier models like Carbon and Argon in multi-model workflows. The architecture offers four thinking levels ranging from low to a new extra-high tier, optimizing output token efficiency. On multi-step tasks such as deep research and codebase navigation, Hydrogen ranks fifth on Agent Arena benchmarks.

The Hydrogen update is currently rolling out to all Google Search users globally. Availability is also expanding to Google's AI Studio, the Gemini API, the Gemini App, and the Antigravity development platform for paid AI Pro and Ultra subscribers.