Gemini 4 Hydrogen model: Google's new fast AI, price and benchmarks

Google released Gemini 4 Hydrogen on Thursday, 12 November. It is the fast, lower-cost member of the new Gemini 4 family, sitting below Carbon and Argon, and it replaces Gemini 3.8 Flash as the model most people will meet in the Gemini app and in Google Search.

“Gemini hydrogen model” was one of India’s top Google searches on Friday morning, and Pakistani developers and freelancers who build on the Gemini API will feel the change too.

What is Gemini 4 Hydrogen?

Google describes Hydrogen as a model for coding, document work and “agentic” tasks, where the AI runs several steps on its own (search, read, write code, check it). Google says it finishes tasks 2.5 times as fast as 3.8 Flash at its default setting.

It reads text, images, audio and video, and has a context window of about one million tokens, so it can take in a long contract or a whole codebase at once. It now thinks harder by default: the standard reasoning level is “high”, and there is a new “xhigh” level for difficult jobs.

Where can you use it?

Google is rolling it out worldwide in the Gemini app, Google Search, the Gemini API, AI Studio and Gemini for business. Free Gemini app users get it automatically when the app picks the model. Google Cloud lists gemini-4-hydrogen as generally available.

How much does Gemini 4 Hydrogen cost?

For developers on the API, the standard price is $1.50 per million input tokens and $7.50 per million output tokens, VentureBeat reported, with the same rate across the full context window. That is twice Flash’s current introductory price, although Flash itself moves up to the same rate on 1 January 2027. Batch and Flex requests cost half.

The per-token price hides part of the story. In independent testing by Artificial Analysis, Hydrogen cost about $1.87 per benchmark task at its high setting, against $7.77 for Flash, because it gets there in fewer steps.

The benchmarks, honestly

• Artificial Analysis Intelligence Index: 51 at high and 54 at xhigh, against Flash’s 31. That puts it level with Google’s bigger Argon model (50).
• Terminal-Bench 4.1 (Google’s figures): 57%, up from Flash’s 18.2%, but still behind Anthropic’s Claude Sonnet 5.5 at 63.9% in Google’s own table.
• Box’s document test: accuracy rose from 74% to 79%, and Box chief executive Aaron Levie said a typical task took less than half the time.

Most of these numbers are Google’s own, so wait for more independent tests before switching a business over.

What it means in Pakistan

For most users the change is invisible: the Gemini app simply gets faster and better at long answers. For freelancers and small software houses that bill clients for AI features, the maths matters more. A model that costs more per token but finishes in fewer steps can be cheaper per job, but only if your tasks are long and multi-step. Simple chat or translation work may cost more than before.

The practical step is to run your own ten or twenty real tasks on both models before the Flash price rise in January, and compare the bill.