Google launches Gemini 4 Hydrogen for agentic workflows

Google has announced Gemini 4 Hydrogen, a new model aimed at everyday coding and agentic workflows. The company says Hydrogen is designed to handle rapid execution while its Gemini 4 Carbon and Gemini 4 Argon models take on longer-horizon reasoning. Google says the model is available across Google AI Studio, the Gemini API, Gemini for business, the Gemini app and Google Search, with paid Google AI Pro and Ultra subscribers also able to use it in Google Antigravity.¹

Hydrogen has a 1-million-token context window, according to Google, with pricing set at $1.50 per million input tokens and $7.50 per million output tokens. Google says the model scored 51 on the Artificial Analysis Intelligence Index at its default high-thinking setting, compared with 50 for Gemini 4 Argon, and reached 54 at its highest xhigh setting.¹

The company is also positioning Hydrogen as the fast-moving part of a multi-model workflow. Google says its Antigravity Orchestrator can switch between models during a session, using Hydrogen for execution while Carbon handles coordination or more demanding decisions. In Google's reported Terminal-Bench 4.0 results, that setup used 24% less quota on average and completed tasks 19% faster than the comparison workflow.¹

Google's own comparisons show Hydrogen completing tasks 2.5 times faster than Gemini 3.8 Flash at the high default setting. The company also reports a rise in Terminal-Bench 4.1 from 18.2% to 57%, alongside gains across coding, knowledge work, computer use, multimodal understanding, science and reasoning. Those figures are Google's claims from its own testing, rather than independent verification, and the benchmark conditions do not by themselves establish how the model will perform across every real-world workload.¹

Availability is the practical part of this announcement. Google says Hydrogen is being offered across several of its consumer, developer and business products, but access can still vary by account, region, plan and rollout timing. Readers using the model for production work should check the relevant product's current documentation and pricing rather than assuming every listed surface has identical limits or features.¹

Our opinion

Gemini 4 Hydrogen looks like Google's clearest attempt to make a fast execution model the everyday workhorse in a larger AI system. The speed and benchmark gains are notable on paper, but the useful test will be whether those claims survive ordinary repositories, failed tool calls and messy multi-step tasks outside Google's selected evaluations. For now, Hydrogen is best understood as a newly announced model with broad promised availability, not a settled verdict on the best agentic workflow.

The announcement matters because Google is selling the workflow around the model as much as the model itself. If Orchestrator can reliably hand deep reasoning to Carbon and routine execution to Hydrogen without wasting context or quota, the arrangement could make agentic development cheaper and easier to manage. That remains a proposition to test, not a result readers should take on trust from one vendor's benchmark table.