Google unveils Gemini 4 Hydrogen, its new AI model built for speed

Google has unveiled Gemini 4 Hydrogen, its newest AI model designed for speed and efficiency. As a successor to Gemini 4 Carbon and Gemini 4 Argon, Hydrogen promises to outperform its predecessors in rapid execution across various agentic workflows. According to Koray Kavukcuoglu, senior vice president at Google DeepMind, Hydrogen is optimized for tasks such as writing code, running tests, and performing everyday work, as opposed to designing complex systems and fixing intricate bugs.

Hydrogen operates on a high thinking level by default, with the option to switch to low, medium, and a new xhigh tier, which differ in quality, latency, and cost. This model is intended to function independently for everyday tasks or collaborate with Carbon and Argon as a high-efficiency execution engine in multi-model workflows. Gemini 4 Hydrogen is set to roll out globally to all users today, replacing Google Search's AI features that were previously powered by Gemini 3.8 Flash.

Google emphasizes that Hydrogen will generate more visual responses and a broader range of relevant links compared to its predecessor. Additionally, it is designed to better comprehend conversational prompts and follow users' thought processes, including handling typos. Kavukcuoglu also mentioned that Hydrogen is less likely to breach permission boundaries compared to Gemini 3.5 Flash by two times, ensuring that users receive more reliable outputs with fewer instances of back-and-forth. However, like all generative AI agents, Hydrogen still faces challenges such as hallucinations.