Google launches Gemini 4 Hydrogen, its fast everyday AI model, in Search and the Gemini app

What did Google launch?

On Thursday, November 12, 2026, Google released Gemini 4 Hydrogen, its fast, lower-cost Gemini 4 model for everyday work. Google DeepMind announced it on X at 1:02pm ET. The launch post pitches it as the quick worker next to its two bigger models, Gemini 4 Carbon and Gemini 4 Argon, which handle the hardest, longest jobs.

Hydrogen is rolling out worldwide starting today. Everyone gets it in Google Search, and it's in the Gemini app's model menu. Free Gemini app users get it when the app's automatic model choice decides it's the right fit for a prompt. Google also points teams still on Gemini 3.8 Flash, its previous fast model, to Hydrogen as the upgrade.

What changes in Google Search?

Google calls Hydrogen "the biggest model upgrade to Search yet," and it's live for all users globally. The Search post lists four changes:

• Messier questions. You can type the way you think, with several parts, typos or very specific needs, and Search should still work out what you mean.
• More links. Google is upgrading "query fan-out," where Search runs many related searches at once behind the scenes, to connect people with "a broader range of helpful links than before."
• More visuals. Expect more images, comparison tables, structured answers and "custom tools built for you."
• More languages. Google says answers get better across all languages AI Mode supports.

What this means for creators and small sites

If Google traffic pays your bills, two of those changes pull in opposite directions. Google says the wider fan-out helps people "find sites you previously might not have discovered," which could send visits to smaller, specific pages. More tables and visuals inside the answer could also mean more people get what they need without clicking.

Neither effect has numbers yet, so watch your own. In Google Search Console, compare clicks and impressions for the weeks before and after November 12. Long, specific questions are the kind Google says Hydrogen handles better. A page that answers one of those plainly, with a clear table where it helps, is the kind of page that could pick up new visits.

Where can you use Hydrogen, and what does it cost?

Tokens are the chunks of text AI is billed by. Input tokens are what you send, output tokens are what the model writes back. Google says Hydrogen's price stays the same across its whole 1 million-token context window, the amount of text it can take in at once, so long documents don't cost more per token.

How fast and smart is it?

These numbers are Google's own, from its launch and Antigravity posts:

• Speed: finishes tasks 2.5 times faster than Gemini 3.8 Flash at its default setting.
• Smarts: scores 51 on the Artificial Analysis Intelligence Index at its default "high" thinking level, ahead of Gemini 4 Argon's 50, and 54 at the top "xhigh" level.
• Thinking levels: four of them, low, medium, high (the default) and xhigh. More thinking means a better answer on hard jobs, but more waiting and a bigger bill.
• Fewer made-up claims: Google says Hydrogen cuts fabricated claims by more than 5x compared with 3.8 Flash.

Google's own figures are a starting point. The real check is how it handles your own work.

What is Orchestrator in Antigravity?

Antigravity is Google's coding app, where AI agents (assistants that carry out multi-step jobs on their own) write and fix code. It needs a paid AI Pro or Ultra plan. The new Orchestrator option, for AI Ultra subscribers only, switches between models inside one job. Carbon plans and untangles the hard parts, and Hydrogen does the fast edits, handing back to Carbon when it gets stuck.

Google says Orchestrator reached near-Carbon success rates while using 24% less quota on average and finishing 19% faster. If you'd rather set your own rules for which model does what, our guide to letting tools pick the model while you cap the bill covers that.

X reactions

Google DeepMind announced the model, and Cognition, the company behind the Devin coding agent, added it to its own tools within the hour.

Meet Gemini 4 Hydrogen: built for rapid coding, multimodal synthesis, and everyday workflows.

It finishes tasks 2.5x faster than 3.8 Flash and surpasses Argon on the @ArtificialAnlys Intelligence Index.

— Google DeepMind (@GoogleDeepMind) November 12, 2026

Gemini 4 Hydrogen is now live in Devin Desktop and Devin CLI.

On FrontierCode 2.0, our benchmark that evaluates models on real-world engineering tasks based on quality and mergeability, Gemini 4 Hydrogen scores 38.2%, just behind GPT-6.1 Sol.

— Cognition (@cognition) November 12, 2026

What to do this week

1. Using Google Search or the free Gemini app? Nothing to do. Hydrogen is already there.
2. Run a site that gets Google traffic? Note your Search Console clicks and impressions for this week, so you have a "before" to compare against later. Our list of SEO tools has free options.
3. Pick one page that answers a long, specific question your readers ask, and make the answer clear in the first few lines.
4. Paying API bills on 3.8 Flash? Run the same real job on Hydrogen and compare the cost per finished job.
5. On Google AI Pro and curious about Antigravity? Hydrogen is in there for you now. Orchestrator needs Ultra.