Google Releases Gemini 4 Hydrogen for Everyday Development
Google has released Gemini 4 Hydrogen, a fast and efficient AI model positioned for rapid execution in coding and agentic workflows. Unlike frontier models like Carbon and Argon, which excel at deep reasoning on complex problems, Hydrogen targets everyday development tasks. The model is now available across Google’s full product suite, from AI Studio and the Gemini API to the consumer Gemini app and Google Search. At $1.50 per million input tokens and $7.50 per million output tokens, Hydrogen completes many tasks 2.5 times faster than Gemini 3.8 Flash while improving reasoning scores more than threefold.
Hydrogen shines in multi-model workflows where it pairs with more powerful models. Google is introducing Orchestrator, which automatically switches between models within a session to optimize cost and speed. The system achieved 24% lower token consumption and 19% faster completion on Terminal-Bench 4.0 while maintaining near-frontier performance. For standalone work, Hydrogen excels at repository coding and debugging with surgical edits, open-ended agent tasks, multimodal knowledge work across documents and video, and web development. It significantly reduces hallucinations compared to Flash and is less likely to overreach in agentic tasks.
The model features four adjustable thinking levels that let users balance quality, latency, and cost for different tasks. In Google’s consumer products, Hydrogen improves instruction-following and context awareness, helping Search understand natural queries better and enabling the Gemini app to complete tasks in fewer iterations. Safety improvements include stronger alignment safeguards and better handling of tool failures.
Read the full article
Continue reading at Hacker News →This is an AI-generated summary. Read the original for the full story.