Google Launches Gemini 4 Hydrogen, a Faster AI Model, in Search and App
Google has released Gemini 4 Hydrogen, a speed-focused AI model now rolling out globally in Google Search and the Gemini app. It replaces Gemini 3.8 Flash in Search and, Google says, is less likely to override permission limits. Google admits it can still hallucinate.
Google unveiled Gemini 4 Hydrogen, a new AI model built for speed, and is rolling it out globally to all users of Google Search and to the Gemini app.
Hydrogen replaces Gemini 3.8 Flash, which has powered Search's AI features since it launched in September. Users can also pick Hydrogen from the model dropdown menu in the Gemini app.
What Hydrogen is for
Google adapted Hydrogen from Gemini 4 Carbon, one of its two frontier models. The other is Gemini 4 Argon. Google claims Hydrogen beats Argon at its default thinking level.
Koray Kavukcuoglu, senior vice president at Google DeepMind, drew a line between the models in the announcement. "While our frontier models, Gemini 4 Carbon and Gemini 4 Argon, excel at deep, long-horizon reasoning on your hardest problems, Hydrogen is built for rapid execution across diverse agentic workflows," he wrote.
In practice, Hydrogen is meant for quick jobs such as writing code, running tests and everyday work. It is not meant for designing systems or fixing complex bugs.
Hydrogen defaults to a "high" thinking level. Developers can switch it to low, medium or a new xhigh tier. The tiers trade off quality, latency and cost.
Kavukcuoglu also said Hydrogen can work alongside the bigger models. "Hydrogen is built to stand on its own as a model for everyday work — or to pair with Carbon and Argon as the high-efficiency execution engine in multi-model workflows," he wrote.
What changes for users
In Search, Google says answers will be more visual and will include a wider variety of relevant links. The model is also designed to follow a user's train of thought in conversational prompts, including when they make typos.
In the Gemini app, Kavukcuoglu said Hydrogen follows instructions more closely across multi-turn conversations. He said it is better at calling the right tools and reaches a reliable output with less back-and-forth.
Safety claims
Google says Hydrogen is two times less likely than 3.8 Flash to circumvent permission boundaries. The company said it is still working on safety and security, including resistance to jailbreaks, which are attempts to trick a model into ignoring its rules.
Google also acknowledged that Hydrogen is still prone to hallucinations, as all generative AI agents are. Those are confident answers that turn out to be wrong.
