Summary
Google has introduced Gemini 4 Hydrogen, a model aimed at coding and agentic workflows, alongside an Antigravity orchestrator that can switch models during a task. Google says Hydrogen offers four reasoning levels and reports faster completion on several coding benchmarks.
Google announced Gemini 4 Hydrogen on Nov 12, 2026, describing it as a fast model for everyday coding and agentic workflows. The company positions it alongside Gemini 4 Carbon and Gemini 4 Argon, which it says are better suited to deep, long-horizon reasoning. Hydrogen is intended for frequent execution work, either on its own or as part of a workflow that assigns different tasks to different models.
A model for execution, with adjustable reasoning
Google says Hydrogen can handle work such as writing code, running tests, debugging and navigating large code repositories. It can also process visual and other files: the company describes support for up to 1 million input tokens and up to 1 million output tokens, including synthesis of PDFs, spreadsheets, charts, audio and recorded video into reports and structured outputs.
The model has four reasoning settings: low, medium, high and xhigh. High is the default. Google presents the settings as a way to trade off reasoning depth, latency and cost according to the task. At the high setting, the company reports a score of 57% on Terminal-Bench 4.1, compared with 18.2% for Gemini 3.8 Flash, and says Hydrogen completes tasks 2.5 times faster in that comparison.
That result concerns a named benchmark rather than every kind of coding work. Google also reports that Hydrogen scored 51 on the Artificial Analysis Intelligence Index at its default high setting, compared with 50 for Argon, and reached 54 at the xhigh setting. These company-reported figures give a picture of performance on particular evaluations; they are not a single measure of how a model will perform across all tasks.
Antigravity can route work between models
Google is introducing Orchestrator in Google Antigravity to switch dynamically between models within a session. The intended approach is to use a model such as Carbon for complex planning and Hydrogen for execution, testing or verification. For work that can be split up, Google says Carbon can coordinate while Hydrogen runs worker and verifier agents in parallel across isolated workspaces.
On Terminal-Bench 4.0, Google reports that Orchestrator achieved task-success rates near Carbon’s while using 24% less quota on average and completing tasks 19% faster. Those are results from the company’s evaluation of this particular multi-model setup.
Google lists Hydrogen as available across all tiers of Google AI Studio, the Gemini API, Gemini for business, the Gemini app and Google Search. It also says paid Google AI Pro and Ultra subscribers can use it in Antigravity. API pricing is $1.50 per million input tokens and $7.50 per million output tokens; Google says the pricing is flat across the model’s 1-million-token context window.