Hydrogen is designed for fast execution across tasks, including workflows in which an AI handles several steps. Google says it completes tasks 2.5 times faster than Gemini 3.8 Flash, its previous fast model. At its default high thinking level, it also beat the more powerful Gemini 4 Argon on the Artificial Analysis Intelligence Index, a broad intelligence benchmark.
In Search, the model is intended to follow natural, detailed questions and address their separate parts. Its reasoning also improves Google’s “fan-out” technique, which breaks a prompt into multiple subqueries to help users find websites they might otherwise miss. In AI Mode, responses may include more images, structured information, comparison tables and custom tools. Google says Hydrogen also improves answers across the languages supported by AI Mode.
Hydrogen has four thinking levels: low, medium, high (the default) and extra-high, or “xhigh.” Google says each is designed to finish tasks faster while using fewer output tokens. The company also says Hydrogen can handle work such as writing code, running tests and checking code changes, either on its own or alongside Gemini 4 Carbon and Gemini 4 Argon. On Agent Arena benchmarks for multi-step tasks such as deep research, data analysis and broad codebase navigation, Google says Hydrogen ranks No. 5.
Beyond Search, Google is bringing the model to all tiers of AI Studio, the Gemini API and Gemini App, as well as paid AI Pro and Ultra subscribers on its Antigravity development platform.
Comments (0)
No comments yet. Be the first!