Tech and AI
Google introduces Gemini 4 Hydrogen, a faster and cheaper model for everyday coding and agent work
Developing
- What happened
- Google has introduced Gemini 4 Hydrogen, a model built for speed and low cost in everyday coding and agent tasks, according to its own announcement.
- Why it matters
- It brings the Gemini 4 generation to all users, including free tiers, and Google says it edges past the larger Argon model on one index.
- What’s next
- It is rolling out across Google’s apps and developer tools, alongside a new Orchestrator mode that switches between models.
Full story
What was announced
Google introduced Gemini 4 Hydrogen on Thursday, the smallest model so far in its Gemini 4 family. Everything known about it at this point comes from Google’s own blog posts; no outside outlet in our sources has tested the claims yet.
According to the announcement, the two frontier models in the family, Gemini 4 Carbon and Gemini 4 Argon, are meant for deep, long-running reasoning on the hardest problems, while Hydrogen is built for quick execution in everyday coding and in agentic workflows, where a model carries out a chain of steps on its own.
Google says Hydrogen brings Gemini 4 to everyone: it is available in all tiers of Google AI Studio, in the Gemini API, in Gemini for business, in the Gemini app and in Google Search, and to paying Google AI Pro and Ultra subscribers in its coding tool Antigravity.
Numbers and price
The company says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default thinking level, against 50 for the larger Argon, and reaches 54 at its highest setting, while often finishing tasks faster than models in its class.
The price is flat across the whole context window of one million tokens: $1.50 per million input tokens and $7.50 per million output tokens. Google describes it as a major upgrade for teams moving from Gemini 3.8 Flash.
Orchestrator
Alongside the model, Google added an option called Orchestrator to Antigravity. Instead of using one model from start to finish, it switches between Carbon and Hydrogen inside the same session. In Google’s example, Carbon first explores the code and makes a plan, Hydrogen writes the edits and runs the tests, and the work is handed back to Carbon if Hydrogen gets stuck.
Google says that on the Terminal-Bench coding benchmark, Orchestrator came close to Carbon’s success rate while using 24% less quota and finishing 19% faster on average. The company contrasts this with a static “model router”, which guesses a task’s difficulty from the first prompt and sends the whole task to one model.
The Antigravity post adds that Carbon was introduced earlier this week together with a feature for running teams of parallel agents, and that tens of thousands of such runs have been launched since.
