Gemini 4 Hydrogen: A new model is now answering your customers

Here is the uncomfortable part. Maybe you checked last month how AI describes your brand. Fine result, box ticked. As of yesterday, that tick is worth less. Not because anything changed about your brand. Because the model talking to your customers changed.

What's new

Google introduced Gemini 4 Hydrogen on November 12. It is pitched as the fast, efficient model for everyday coding and agent workflows. Google's numbers:

• API price: $1.50 per million input tokens and $7.50 per million output tokens, flat across the full 1 million-token context window.
• On the Artificial Analysis Intelligence Index, Hydrogen scores 51 versus 50 for the larger Argon at the default thinking level, and 54 at the highest setting.
• Compared with Gemini 3.8 Flash, Google says fabricated claims drop by more than 5x.
• Available in the Gemini API and Google AI Studio, in the Gemini app model menu, and in Google Search for all users. Free app users get it through automatic model selection when it fits the prompt.

Two days earlier Google shipped Gemini 4 Carbon and told most users to stick with Argon. That advice has already aged. This lineup moves faster than most procurement cycles.

Search is the channel, not the model

Most coverage will be about benchmarks and token prices. For marketers that is a footnote. The sentence that matters in Google's post: Hydrogen significantly upgrades Search for all users globally. Google does not say which Search feature. What is clear is that some of the answers your customers read on Google now come from a different model.

Which means AI visibility is not a state you reach once. It is a moving number that can shift with every model swap, without you doing anything. And free users no longer even pick their model. Google routes for them.

The fabrication line is the one I find most revealing. If the new model makes up more than five times fewer claims than its predecessor, the predecessor made up more. Possibly about you. Whether the new answers are better, you only know once you look.

The catch

Every number comes from Google or from early-access partners. And Google gives no per-country timeline and nothing EU-specific, so what a user in Munich or Chicago sees today is not spelled out.

What I would do now

My rule of thumb: the question is no longer just where you rank on Google, but how the AI describes your brand, and whether you make the relevant set at all when someone asks without your name.

• Take five real customer questions without your brand name and ask them in Google and the Gemini app today. Log date, answer and cited sources.
• Compare against your last baseline. New competitors in the answer? Wrong facts gone, or new ones in?
• Make this a fixed check after every major model release, not once a quarter.

Bottom lineGoogle quietly swapped the salesperson at your shelf yesterday, and nobody told you.