Google released Gemini 4 Hydrogen on 12 November 2026 through the Gemini API and Google AI Studio. The model is aimed at coding, tool use and other recurring agent tasks, while Google positions Gemini 4 Argon and Carbon for more demanding reasoning. For developers, the concrete questions are how much context Hydrogen can handle, what its API calls cost and whether its speed holds up in their own workflows. Google's launch announcement describes the release and its intended use.
The developer terms
The official API pricing table lists the model as gemini-4-hydrogen with a unified one-million-token context window. Standard paid pricing is $1.50 per million input tokens and $7.50 per million output tokens. Google says the same rates apply across that window rather than changing at a long-context threshold. The table also lists a free tier; teams should check its current limits before treating free access as production capacity.
Google offers four thinking levels: low, medium, high and xhigh, with high as the default. This gives developers a way to compare answer quality, latency and token use on the same task instead of assuming the highest setting is always the best fit. The developer release note says Hydrogen can run on its own or alongside other Gemini 4 models.
Model routing is a separate product feature
Google also introduced Orchestrator in Antigravity. It can switch between models such as Carbon and Hydrogen during a session, assigning more demanding work to one and faster execution to another. This is an Antigravity workflow, not a promise that a direct Hydrogen API request will choose another model automatically. Antigravity access to Hydrogen is described for paid Google AI Pro and Ultra subscribers, a separate condition from the API pricing table.
Google reports benchmark and partner-test gains for Hydrogen over older models, including faster task completion and better coding results. Those figures come from Google's release material and selected evaluations. They do not establish the same improvement for every repository, prompt or tool environment. Teams considering migration should test representative tasks, inspect code changes and failures, and compare total cost per completed task rather than token price alone.
The release expands the available choices for agent systems: a lower-priced model with adjustable reasoning, a large unified context window and optional product-level model routing. Its operational value will depend on measured reliability and latency in each team's own workload, particularly when the model can call tools or modify code.
