Google introduced Gemini 4 Hydrogen on November 12, targeting everyday coding and tool-using agents. Available through Google AI Studio and the Gemini API, it offers a unified one-million-token context window. Paid API pricing is $1.50 per million input tokens and $7.50 per million output tokens. The Google announcement positions Hydrogen for rapid execution alongside its larger reasoning models.
The developer guide makes an important context distinction: input, thinking and output share the same window. It is not a separate one-million-token allowance for each.
Orchestrator: a planner and an execution model
The accompanying Antigravity Orchestrator changes how models work together. Instead of choosing one model for an entire task, it switches between a larger planner and Hydrogen: plan, implement, test, then return to the planner when execution gets stuck. These handoffs happen within the task, rather than simply routing the initial request to one model.
The Carbon variant supports Google AI Ultra subscribers; the Argon variant also supports paid Google AI Pro subscribers. The Antigravity launch and current model-selection documentation describe the planner and execution roles and plan-specific availability.
Why it matters for builders
The practical shift is from choosing one coding model to assigning different models different jobs.
For small teams such as SunMarc App Labs, that suggests a useful evaluation: keep architecture and difficult debugging with a stronger planner, then test Hydrogen on bounded app fixes, website changes and verification. Compare completed-task cost, elapsed time and regressions—not just token prices.
Google reports faster, more quota-efficient execution with Orchestrator. Those are vendor results, not a SunMarc hands-on test. The release alone does not establish savings on our own projects.
For more context on Google's larger model and agent workflows, see our Gemini 4 Carbon coverage.