Google Launches Gemini 4 Hydrogen, Calling It the Biggest Search Upgrade Yet

admin · November 12, 2026 · 7 min read

Illustration of Google's Gemini 4 Hydrogen AI model powering a colorful Google Search results page

Google Launches Gemini 4 Hydrogen, Calling It the Biggest Search Upgrade Yet

Google has released Gemini 4 Hydrogen, a new artificial intelligence model that the company says delivers the biggest upgrade to Google Search yet. The model went live on Thursday, November 12, and is already running in Search for users around the world, alongside a rollout across Google’s developer tools and the Gemini app.

Google announced the model in a blog post from Google DeepMind on Thursday afternoon. “Today, we’re introducing Gemini 4 Hydrogen, bringing cutting-edge intelligence to everyone — with speed and efficiency for everyday coding and agentic workflows,” the company wrote. For Search, Google said Hydrogen is “significantly upgrading” the experience, “delivering more helpful, intuitive responses for all users globally.”

The launch arrives just days after Google released Gemini 4 Carbon for AI Mode, a version available only to paid subscribers. Hydrogen takes the opposite approach: it is the default model in Search for everyone, free of charge, and it is also available to developers through the Gemini API and Google AI Studio on all tiers.

Faster, cheaper, and built for everyday work

Hydrogen is positioned as the successor to Gemini 3.8 Flash, Google’s previous fast, low-cost model. According to Google, it completes tasks 2.5 times faster than Gemini 3.8 Flash at its default thinking level, and its score on Terminal-Bench 4.1, a benchmark for agentic coding, jumped to 57 percent from 18.2 percent.

The model offers a one-million-token context window and can produce up to one million output tokens through background long-form generation, with up to 64,000 tokens in a standard synchronous response. It comes with four adjustable thinking levels — low, medium, high and xhigh — with high set as the default, letting users trade speed for deeper reasoning when a task demands it.

On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default level, one point ahead of Gemini 4 Argon, and 54 at its highest thinking level.

Pricing is aimed squarely at high-volume use. Through the Gemini API, Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, and Google offers a free tier. It is also included in Gemini for business and in Google Antigravity, the company’s agentic development environment, for paid Google AI Pro and Ultra subscribers.

What changes in Google Search

For most people, the most visible change will be in Search itself. Google said Hydrogen is now available globally in Search for all users, and it outlined four improvements the model brings to everyday queries.

First, Search can now handle more nuanced and complicated questions, including multi-part questions, hard-to-understand typos and highly specific requests. With deeper instruction following and tool calling, Google said, Search follows what users want more precisely and addresses each part of a question.

Second, Hydrogen’s reasoning improves Search’s query fan-out technique, running simultaneous queries across the relevant subtopics of a question and connecting users to a broader range of links than before — including sites they might not otherwise have discovered.

Third, responses are becoming more visual and easier to scan, with more images, structured responses, comparison tables and custom tools built for the query. Fourth, Google said language understanding is richer across all supported AI Mode languages, no matter which language a user asks in.

The company also published a model card for Hydrogen, which notes the model’s limits alongside its strengths. It lists hallucinations, occasional slowness or timeouts, and potentially greater token use at higher thinking levels, and it gives a March 2026 knowledge cutoff, warning that knowledge in some domains may be less current.

Background: from leaked codename to official launch

Hydrogen was one of the worst-kept secrets in AI this autumn. For weeks, testers on social media reported sightings of a model checkpoint labelled Hydrogen in Google AI Studio and on public evaluation leaderboards, and many speculated it was the unannounced Gemini 4 Flash. Google had not confirmed any of those sightings, and its public model list showed only Gemini 4 Argon and Gemini 3.8 Flash.

Thursday’s announcement settles the question. Google’s three launch posts describe Hydrogen as the upgrade for teams moving from Gemini 3.8 Flash, and they never use the name Gemini 4 Flash. The model is now listed in Google’s API changelog with an entry dated November 12, 2026.

The release also lands in the middle of a crowded week for AI competition. Anthropic withdrew IPO pricing details from banks earlier this month after a filing valuing the company at nearly $70 billion circulated, and the AI industry’s biggest names are racing to ship faster, cheaper models that can run autonomous coding and research agents at scale. Read more of our reporting in the AI section.

Why this matters

Hydrogen matters because of where Google put it: not behind a paywall or a waitlist, but inside the search box billions of people already use. Every Google Search user now gets responses powered by the new model by default, which makes it one of the largest AI deployments ever attempted in a single day.

For developers and businesses, the combination of low pricing, a one-million-token context window and a free API tier lowers the cost of building agents that can read large codebases, research the web and carry out multi-step tasks. For publishers and website owners, the change could reshape how their pages surface in Search, since Hydrogen now drives the AI Mode and AI Overview responses that sit above traditional results.

Google’s own framing is ambitious: the company is betting that a faster, cheaper model used by everyone will do more for its AI position than a flagship model reserved for paying customers. Whether Hydrogen lives up to that billing will become clear quickly, because unlike most AI launches, this one is already in front of the whole world.

Latest developments

Google said Hydrogen is rolling out across the Gemini API, AI Studio, the Gemini app, Gemini for business and Search, with Antigravity access for paid Pro and Ultra subscribers. The company has not announced a separate launch event or additional pricing tiers, and its developer documentation was updated on Thursday with the model’s pricing, context window and benchmark figures.

Frequently Asked Questions

What is Gemini 4 Hydrogen?

Gemini 4 Hydrogen is Google’s newest AI model, launched on November 12, 2026. It is designed for everyday coding and agentic workflows, and it now powers Google Search for all users globally, as well as the Gemini API, AI Studio and the Gemini app.

How much does Gemini 4 Hydrogen cost?

Through the Gemini API, Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens. Google also offers a free tier on the API, and the model is included in Gemini for business and in Google Antigravity for paid Google AI Pro and Ultra subscribers.

Is Gemini 4 Hydrogen available in Google Search?

Yes. Google said Hydrogen is available globally in Google Search for all users, calling it the biggest model upgrade to Search yet. It brings better understanding of complex questions, improved query fan-out, more visual responses and richer language understanding.

How is Hydrogen different from Gemini 3.8 Flash?

Google says Hydrogen completes tasks 2.5 times faster than Gemini 3.8 Flash at its default thinking level, and its Terminal-Bench 4.1 score rose to 57 percent from 18.2 percent. It also offers a one-million-token context window and four adjustable thinking levels.

Final paragraph

Google’s launch of Gemini 4 Hydrogen puts a faster, cheaper AI model directly into the world’s most-used search engine on day one. The benchmarks and the pricing suggest a model built for scale rather than spectacle, and its real test will be whether the billions of searches it now powers actually feel smarter. Readers should watch for independent benchmark results and for how publishers and rival AI labs respond in the coming weeks.

Sources: