Google introduced Gemini 4 Hydrogen on November 12 as a model for everyday coding and agentic workflows. The company positions Hydrogen as a fast execution model alongside Gemini 4 Carbon and Argon, which it describes as better suited to difficult, long-horizon reasoning. Google says Hydrogen is rolling out globally through the Gemini API, AI Studio, Gemini for business, the Gemini app and Search, with access on some surfaces depending on subscription or eligibility.
Hydrogen supports a one-million-token context window and four thinking levels: low, medium, high and xhigh. Google lists API prices of $1.50 per million input tokens and $7.50 per million output tokens across the context window. The price and availability details matter because the model is being positioned for frequent execution work, where latency and usage cost can shape whether teams delegate tasks to an agent.
Google also describes Orchestrator in Antigravity, which can switch between models during a session, using Carbon for coordination and Hydrogen for worker or verifier tasks. On Terminal-Bench 4.0, Google reports that Orchestrator approached Carbon's task-success rate while using 24% less quota and finishing 19% faster on average. Those are issuer-reported evaluation results, not a guarantee of equivalent savings for every workload.
For coding, Google reports a 57% score on Terminal-Bench 4.1 for Hydrogen, compared with 18.2% for Gemini 3.8 Flash, and says Hydrogen completed tasks 2.5 times faster at its default high setting. Google also reports lower fabricated-claim rates and improved recovery from failed tool calls. These comparisons depend on Google's stated benchmark setup and should be checked against independent evaluations and production use.
The launch reflects a model strategy that pairs a top-end system with a faster workhorse and lets software route between them. Whether that reduces total cost or improves completion rates will depend on the task, the routing policy and the customer's access tier; Google's benchmark and partner examples are early vendor evidence rather than independent proof of broad performance.