Los Angeles, CA
Sections
The LA Globe
AI

Google's Gemini 4 Hydrogen Now Writes AI Search Answers for Everyone

The faster model replaces Gemini 3.8 Flash. In the Gemini app, Ars Technica reports, only Google AI Pro and Ultra subscribers get it outright.

2 min read
ShareXFacebookEmail
Comments
Library patrons seen in profile typing at a row of public computers under fluorescent light.
Photo illustration: Patrons at a public library's computer row. (Photo illustration: The LA Globe)

Type a rambling, multipart question into Google, typos and all, and a new AI model is now writing the answer. On Thursday Google said it had put Gemini 4 Hydrogen into Search for all users worldwide, and it called the switch the biggest model upgrade Search has ever had.

Hydrogen is Google's fast, cheaper tier. It replaces Gemini 3.8 Flash, the model most people actually use across Google's AI products, Ars Technica reported. It also skipped the slow rollout Google gave its premium Carbon and Argon models. Those went to cybersecurity researchers first, and then only to subscribers on the company's most expensive AI plan.

In Search, Google says, Hydrogen should untangle long questions with several parts, answer with more images and comparison tables, and point to a wider spread of websites. Google's DeepMind announcement claims the model makes up facts more than five times less often than 3.8 Flash did, a number from Google's own testing. Ars noted that Google has promised more links before.

In the Gemini app, "everyone" gets smaller. Google's announcement says Hydrogen brings Gemini 4 "to everyone," but Ars reported that in the app it's live for Google AI Pro and Ultra subscribers, while AI Plus and free users are still on the older Gemini 3.5 Flash-Lite. A footnote on Google's own page says free users get Hydrogen only through automatic model selection, when the system decides it's the right model for the prompt.

Developers who build on Hydrogen pay what they paid for 3.8 Flash. Ars Technica reported that Hydrogen's API rate matches 3.8 Flash's: $1.50 for each million tokens sent in, and $7.50 for each million the model sends back. Carbon and Argon currently run $2 and $10 on the same terms, and Ars reported those rates double to $4 and $20 in December.

Google also says that at its default setting, Hydrogen scored 51 on the Artificial Analysis Intelligence Index, a point above Argon's 50. By Google's own scorecard, then, the model with the unchanged price beats one whose price is about to double.

Source: arstechnica.com, retrieved November 13, 2026. Other sources: blog.google, deepmind.google.

Have a tip we should know? Email tips@thelaglobe.com.

Stay in the Orbit

Two emails a week, Mondays and Fridays: the stories worth your time. Free. Unsubscribe any time. Privacy.

Or follow the Globe on Facebook.

Comments