Developers can use Gemini 4 Hydrogen from Google on AI Gateway

Gemini 4 Hydrogen is on AI Gateway, which gives one API for many models. It replaces Gemini 3.8 Flash and stays in the same latency class. The text tells of more skill in agentic code work, knowledge work and tool use. It gives no number and no test result. The model is for developers who make code agents.

What is new

Hydrogen follows system instructions and task limits more carefully. It controls tools more carefully in long chats. It makes documents, spreadsheets and presentations with a lower number of agent steps.

The text also tells of more grounding. It cuts claims with no proof in search answers and false test results in code reports.

How to use it

Use google/gemini-4-hydrogen as the model name. To use it in a code agent, install the new Vercel CLI and do the setup. Then select the model in the agent.

AI Gateway also shows use and cost, and has budgets for API keys.

What is not known

The text gives no test result for each claim. It does not tell the price of the model.

Insight: A gateway with no fee favors the lowest cost

Google tells that Hydrogen has 57% on a coding test, against 18.2% for the earlier model. The price is $1.50 for 1,000,000 input tokens. AI Gateway adds no markup and no fee. This reads as a contest to be the usual route. With no fee at the gate, the model with the least cost for each task is the one that teams select. Compare cost for each task, not the price.