Google has released Gemini 4 Hydrogen, a faster and cheaper model built for everyday coding and agent work. It sits below Google’s frontier models, Gemini 4 Carbon and Gemini 4 Argon, but Google says it already outscores Argon on an independent index.
It also reaches far more people. Free users of the Gemini app get Hydrogen through automatic model selection, and Google is bringing it to Search for all users.
What Google released
In its launch post, Google DeepMind says Hydrogen is built for rapid execution across agent workflows. It is meant for the steady work of an agent, such as running steps, checking results and handling routine coding, rather than the longest reasoning jobs.
Google also pairs it with its bigger models. In Google Antigravity, a new Orchestrator can switch between Carbon and Hydrogen in one session, with Carbon coordinating parallel agents and Hydrogen doing the worker and checking tasks.
Who can use it, and where
Free users get Hydrogen in the Gemini app, but not as a manual choice: automatic model selection decides when to use it for each prompt. Google AI Pro and Ultra subscribers can also use it in Google Antigravity and in Gemini in Google Sheets. The jump from free to Pro is where the price lands, and our guide to Gemini plans in Australia lists each one in local dollars.
Developers can use it through the Gemini API on the paid tier in Google AI Studio, and in Android Studio. Businesses get it through Gemini for business. Google calls the rollout global but does not list countries, so Australian users should see it as it arrives.
What it costs on the API
Hydrogen costs $2.19 per million input tokens and $10.96 per million output tokens, and Google keeps that price flat across its full one million token context window. Google bills in US dollars, so the local cost moves with the exchange rate.
That puts it below Claude Sonnet 5.5 and GPT 6.1 Sol, which both cost $2.92 in and $14.62 out per million tokens. For agents that run thousands of small steps, that gap adds up quickly.

How it compares, by Google’s numbers
Google says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high thinking level, one point ahead of Argon, and 54 at its highest setting. Against Gemini 3.8 Flash, Google reports 57 per cent on Terminal Bench 4.1 against 18.2 per cent, and says Hydrogen finishes tasks 2.5 times faster.
These are Google’s own figures, and independent speed tests are not out yet. Our comparison of Gemini 4 Argon and Claude Opus 5.5 shows where Google’s bigger model stands against Anthropic’s.
Why Australian businesses should care
For most small teams, the change is quiet: if you use the free Gemini app, some of your prompts will now be answered by a stronger model without you choosing it. The bigger shift is for anyone building agents, where a capable model at this price can change whether an automation project pays for itself.
If you are weighing Gemini against other assistants, our guide to Claude and Gemini sets out where each one fits.
Worth testing on real work
Hydrogen is a sign that the cheaper tier of models keeps catching up with last season’s flagships. The sensible test is to run it on a sample of your own tasks next to your current model and compare the bill and the error rate.
It follows Gemini 4 Carbon, which Google launched in October and which reached GitHub Copilot this week. Between the two, Google now has a model for the hardest problems and another for the volume of everyday agent work.
