Google Launches Gemini 4 Hydrogen: A Fast Model for AI Agents

· Cortex Solutions

Conceptual illustration of a glowing fast AI core connected to a spreadsheet, code window, search bar and chat bubble

Google launched Gemini 4 Hydrogen on Nov 12, 2026: a fast agentic model, 2.5x faster than 3.8 Flash per Google, priced at $1.50/$7.50 and now in Search.

On November 12, 2026, Google announced Gemini 4 Hydrogen, a fast model it presents as built for "agentic workflows" and everyday coding, meaning tasks where AI carries out a sequence of steps on its own rather than just answering one question. According to Google's official post by Koray Kavukcuoglu, SVP at Google DeepMind, the model is rolling out globally from the day of the announcement in the Gemini API, the Gemini app, Google Search and Gemini for business.

Where Hydrogen sits in the Gemini 4 family

Google places Hydrogen below the two larger models in the family, Gemini 4 Carbon and Gemini 4 Argon, which it describes as being for deep, long-horizon reasoning. Hydrogen, the announcement says, is "built for rapid execution across diverse agentic workflows": the model meant for daily, repeated tasks.

According to Google, Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at its default setting. The company says early-access partners testing it on graphs, SQL and data-heavy workflows found it 2x to 2.8x faster, with 12% to 23% lower token use. Google also says its outputs in Google Sheets are more than 30% faster than previous models.

The model has four configurable thinking levels: low, medium, high (the default) and a new, higher level Google calls xhigh.

What the numbers say

The main figures published by Google and reported by VentureBeat on the same day:

  • Artificial Analysis Intelligence Index: 51 at the default setting and 54 at the highest, compared with 31 for Gemini 3.8 Flash according to VentureBeat.
  • Terminal-Bench 4.1 for command-line coding tasks: 57% versus 18.2% for the previous model, though VentureBeat noted it still trails Claude Sonnet 5.5 at 63.9% in Google's own table.
  • Specialised tests from the evaluator Vals cited by Google: 52.3% on a tax agent benchmark and 55.8% on legal research.

Most of these figures come from Google itself or are reported from it, and independent evaluation is still at an early stage.

Price: faster, but not the cheapest

Google set Hydrogen's API price at $1.50 per million input tokens and $7.50 per million output tokens, flat across its full one-million-token context window. VentureBeat added that other tiers, such as batch processing, cost half the standard rate in exchange for slower turnaround.

VentureBeat's headline called it a "faster, smarter, pricier" Flash successor, and said its standard price is about 15 times the lowest standard rates of smaller rivals such as Claude Haiku 5.5 and GPT-6 Luna. It also noted that raising the thinking level to xhigh did not improve the result on one benchmark, while the estimated cost rose from $1.87 to $3.16. In short, choosing the right thinking level for each task will directly affect the bill.

Hydrogen in Google Search

In a separate post on November 12, Robby Stein, VP of Product for Google Search, called it "the biggest model upgrade to Search yet" and said Hydrogen is available to all users globally. According to Google, the upgrade includes:

  • Better understanding of user intent in multi-part questions, even with typos.
  • A broader range of links in answers.
  • More visual answers with images and comparison tables.
  • Richer language understanding across all languages supported by AI Mode.

Where the model is available

According to Google, Hydrogen is available to developers through the Gemini API, Google AI Studio and Android Studio; to organisations through Gemini for business and Gemini in Google Sheets for eligible business subscriptions; and to individuals in the Gemini app. Free users reach it through automatic model selection when it suits the prompt. Google did not publish a list of supported countries or languages in the announcement, saying only that it is rolling out globally.

What it means for Saudi businesses

  • AI is moving from answering to doing: fast models built for agentic workflows bring automation of repetitive tasks, such as summarising documents, analysing spreadsheets and first-line customer replies, closer to daily use.
  • Cost needs calculating: the model is faster but not the cheapest, so work out the cost per task before scaling, and match the thinking level to how much the task matters.
  • Don't rely on vendor numbers alone: test the model on your own data and real tasks in Arabic before deciding.
  • Search is changing: more detailed search answers may change how customers reach your website, so keep your content clear and useful.
  • Start small with human oversight: pick one process, put human review on its outputs, then expand once you have measured the results.

If you want to turn these models into practical tools inside your business, the Cortex team can help you design and run AI automation built around your work: https://cortex.sa/services/ai-automation

Sources: Google's official posts dated November 12, 2026 (model announcement, Search upgrade and developer blog), and VentureBeat's report dated November 12, 2026.

Read next