Google's Gemini 4 Hydrogen is rolling into Search and the app

Anyone using Google Search or the Gemini app may already be talking to a new model without changing a setting. Google introduced Gemini 4 Hydrogen on Nov. 12 as the replacement for Gemini 3.8 Flash, and it decides on its own when to use it.

Free users of the Gemini app get Hydrogen when the automatic model picker judges a question worth it. Google says most free queries still run on Flash-Lite, its smallest model. In Search, the company calls it the biggest model upgrade yet, and says it handles multi-part questions, typos and very specific requests better, with more images and comparison tables in answers.

The announcement names no countries, Mexico included, and describes a global rollout starting Nov. 12. It does not say whether languages, plans or timing differ for Spanish speakers.

What developers will pay

Hydrogen is built for coding and for agents, programs that chain tasks and tools together on their own. Developers can find it in AI Studio and the Gemini API, and it also ships in Android Studio and Gemini for enterprise. Antigravity, Google's agentic coding environment, offers it to AI Pro and Ultra subscribers.

Source: Google

A token is a chunk of text, roughly a syllable or a short word. Google says the price stays flat across the whole context window. The API calls the model `gemini-4-hydrogen`, and developers can choose how much it "thinks" before answering. There are four levels: low, medium, high (the default) and a new one, "xhigh".

The numbers are Google's

Every result in the announcement comes from the company. Among them:

• Tasks finish 2.5 times faster than on its predecessor, Gemini 3.8 Flash, at the high setting. In partner tests the gain was 2 to 2.8 times, using 12% to 23% fewer tokens.
• It scores 57% on Terminal-Bench 4.1, against 18.2% for 3.8 Flash.
• It reaches 51 on the Artificial Analysis Intelligence Index at high, against 50 for Gemini 4 Argon, and 54 at xhigh.
• On safety, Google reports up to 81% less overreach by agents than 3.8 Flash, and more than 2.3 times the resistance to indirect prompt injection.

Artificial Analysis is an outside evaluator, but the figures in the announcement are Google's presentation, and nobody else has reproduced them. The remaining benchmarks are the ones the company chose to publish.

Koray Kavukcuoglu, Google's chief AI architect, wrote that Hydrogen is built for fast execution across agent workflows. The official @GoogleAI account said it "brings cutting-edge intelligence to everyone."

Hydrogen sits below Gemini 4 Carbon and Argon, which Google keeps for long reasoning and hard problems. For anyone paying for the API, the missing test is their own: what it costs and how fast it runs on their workload, set against the announcement's figures.