Google has released Gemini 4 Hydrogen, a flash-class model designed for speed and efficiency in coding and agentic workflows. The new model replaces Gemini 3.8 Flash and is available across Google AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search, and Google Antigravity for paid Pro and Ultra subscribers.
Google CEO Sundar Pichai stated that the model is "built for speed and efficiency for everyday tasks, coding and agentic workflows." According to Pichai, Hydrogen completes tasks 2.5 times faster than 3.8 Flash and features four distinct thinking levels to allow users to select the necessary reasoning depth.
The model is priced at $1.50 per million input tokens and $7.50 per million output tokens across its 1 million-token context window. Vals AI identified Hydrogen as the cheapest and fastest model among its top six ranked models. At the xhigh reasoning level, the model's cost is approximately 32 percent of the cost of Claude Sonnet 5.5 (max) for similar tasks, though it costs roughly 2.5 times as much as GPT-6.1 Sol (max).
In performance benchmarks, Hydrogen scored 54 on the Artificial Analysis Intelligence Index at its highest thinking setting, surpassing Gemini 4 Argon's score of 50. The model also achieved a 57% score on Terminal-Bench 4.1, more than tripling the 18.2% score of 3.8 Flash. It currently ranks fifth on the Agent Arena benchmark and leads its class on the HLE-Diamond and AutomationBench-AA benchmarks.
Google is also introducing Orchestrator in Google Antigravity to facilitate multi-model collaboration. This feature allows the system to dynamically switch between models, such as Gemini 4 Carbon and Hydrogen, within a single session. On Terminal-Bench 4.0, Orchestrator reportedly reached near-Carbon task success rates while consuming 24% less quota and completing tasks 19% faster.
