Google brought Gemini 4 Hydrogen to Search users worldwide on 12 November, according to a release by Robby Stein, its vice-president of Search product. Google says the model can handle complicated questions with several parts and give more visual answers, extending the upgrade to all languages supported by its AI models.
Key points
- Google says Gemini 4 Hydrogen is available in Search globally for all users.
- The Search upgrade is intended to handle multi-part questions and return images, tables and relevant links.
- Google puts Hydrogen at 51 on the Artificial Analysis Intelligence Index at its default high thinking setting, according to 9to5Google.
- The Gemini app lists Hydrogen for “Everyday help”.
Gemini 4 Hydrogen changes the Search answer
Google says Hydrogen follows detailed requests more closely, including questions with several parts, unusual typos and highly specific requirements. Stein also says its reasoning connects searches to a wider range of relevant websites. In the release, Google describes responses that bring together written answers, images, comparison tables and links, alongside custom tools.
The distinction for Search is between retrieving material relevant to the words in a request and responding to the requirements those words express together. A comparison table, for example, puts details side by side rather than leaving them scattered through an answer. Google presents that combination of intent-following and structured output as the purpose of the Hydrogen upgrade, including for questions phrased in a natural flow of thought.
Comparing purchases with several requirements could therefore produce a table alongside relevant links, if Search handles the request as Google describes. The table would make the options easier to inspect together, while the links would provide a route to the websites behind the answer. That remains a use of the capabilities Google describes, rather than a measured outcome of the rollout.
The worldwide release puts those capabilities in ordinary Search, rather than requiring a choice from the Gemini app’s model picker. Google says the updated responses cover every language its AI models support. For sites and services that rely on being found through Search, the described change is also a change in presentation: Google says its answers can place links within a more structured response to a detailed query.
Hydrogen scores 51 at high thinking
Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high thinking setting, against 50 for Gemini 4 Argon, according to a Google claim quoted by 9to5Google. The publication also reports Google’s figure of 54 for Hydrogen at its highest, xhigh, thinking setting. The scores concern those stated settings on the index, not a measurement of Search answers delivered to users.
A thinking setting controls how much work a model is allowed to put into a response before producing it. That matters when reading the two Hydrogen figures: 51 is the score Google gives for the default high setting, while 54 belongs to a different, more intensive setting. Google also says Hydrogen often finishes tasks faster than models in its class, a claim reported alongside the index figures.
The benchmark comparison gives builders choosing between models a result for particular configurations. Search presents a different job: Google’s release describes questions, links and visual answers, whereas the index figures reported by 9to5Google are model scores. A choice about an application would depend on the setting used and the work it needs done, rather than treating the highest quoted score as the Search experience.
Hydrogen takes the everyday-help position in Gemini
Google has also put Hydrogen in the Gemini app’s model picker under “Everyday help”, 9to5Google reported. The picker assigns quick answers to 3.5 Flash-Lite, code and analysis to 4 Argon, and the most complex work to 4 Carbon. Hydrogen’s placement gives the new name a defined role alongside those models, rather than making it a replacement for every option in the app.
The naming marks a change from Gemini 3.8 Flash, which launched on 2 September. Google introduced Hydrogen in place of continuing the Flash branding, after its Argon and Carbon models. AI Affairs previously covered Carbon’s wider release and Google’s preview of agent swarms. Hydrogen’s Search deployment puts a differently positioned model into a service available to all Search users.
The division extends into Google’s building tools. Google Antigravity offers Hydrogen an “Extra” thinking tier, according to 9to5Google, while its Orchestrator description says Carbon may hand routine steps to Hydrogen to use fewer tokens. That arrangement describes a possible way to divide work between models, with the more demanding task kept with Carbon and routine steps assigned elsewhere.
