Google has introduced Gemini 4 Hydrogen, a new model aimed at software development, enterprise processes and everyday consumer tasks. The company describes it as an execution-focused addition to the Gemini 4 lineup, designed to handle coding, debugging and multi-step workflows without relying on the heavier models in the family for every task.
The model joins Gemini 4 Carbon and Gemini 4 Argon. Google positions Carbon and Argon for more demanding reasoning and extended tasks, while Hydrogen is intended to complete routine work quickly and efficiently, either on its own or alongside the other models. The announcement was made by Koray Kavukcuoglu, Google DeepMind’s senior vice president and chief AI architect.
Pricing and performance claims
Google has set Gemini 4 Hydrogen’s API price at $1.50 per million input tokens and $7.50 per million output tokens. Those rates apply across the model’s one-million-token context window, according to the company.
Google says Hydrogen completes certain tasks 2.5 times faster than Gemini 3.8 Flash when using its default high-thinking setting. The company also reported a rise in its Terminal-Bench 4.1 score from 18.2% to 57%, more than tripling the earlier result. The benchmark figures are company-reported performance claims rather than an independent assessment.
| Item | Gemini 4 Hydrogen |
|---|---|
| Input API price | $1.50 per million tokens |
| Output API price | $7.50 per million tokens |
| Context window | One million tokens |
| Reported speed versus Gemini 3.8 Flash | 2.5 times faster on certain tasks |
| Terminal-Bench 4.1 score | 57%, compared with 18.2% for Gemini 3.8 Flash |
Designed for agentic software work
The launch is focused in part on developers working with large codebases. Google says Hydrogen is better suited to debugging applications, running tests and coordinating several tasks through AI agents. It also reported that the model cuts fabricated claims by more than five times compared with Gemini 3.8 Flash, while improving its ability to recover when a tool call or search fails.
Those capabilities are intended to address practical problems in software workflows, where an AI system may need to interpret a large repository, use external tools and continue after an unsuccessful step. Google’s description presents Hydrogen as a model for repeated execution and operational speed rather than as the highest-capability option in the Gemini 4 range.
Models working together
Google is also introducing Orchestrator, a capability within Google Antigravity that allows multiple Gemini models to collaborate in one workflow. The arrangement is designed to let different models handle different stages of a task instead of forcing one model to perform every step.
- Gemini 4 Hydrogen is aimed at fast execution, coding and debugging.
- Gemini 4 Carbon and Argon are positioned for complex reasoning and longer-running work.
- Orchestrator allows models to collaborate within a shared workflow in Google Antigravity.
Google’s launch expands its effort to offer models with different balances of speed, capability and operating cost. Hydrogen’s pricing and reported performance are aimed at organizations running frequent software and AI-agent tasks, while the model’s place alongside Carbon and Argon gives developers more than one option for assigning work across the Gemini 4 family.