Google Introduces Gemini 4 Hydrogen as a Fast and Cost Effective Model
November 15, 2026 · 2 min read
Also published in Português (Brasil)
Google has released Gemini 4 Hydrogen, an artificial intelligence model designed as a fast all rounder for end users and developers.
Hydrogen is intended for everyday tasks and is positioned as a more affordable alternative to the Gemini 4 Carbon and Gemini 4 Argon models, which are meant for deep reasoning on difficult problems. According to Google, Hydrogen is a major upgrade over Gemini 3.8 Flash, completing tasks 2.5 times faster at the standard reasoning level. Its score on Terminal Bench 4.1 rose from 18.2 percent for Flash to 57 percent for Hydrogen, and hallucinations are reportedly reduced by more than five times.
The model features four reasoning levels: low, medium, high, and xhigh. On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at the high level and up to 54 at the xhigh level, which surpasses the Argon model's score of 50 at the high level. Hydrogen supports a context window of one million tokens, with a price of 1.50 US dollars per million input tokens and 7.50 US dollars per million output tokens. The cache hit cost is 0.075 US dollars.
In multimodal knowledge work, Hydrogen supports up to one million output tokens in addition to its input window. It achieved 53.4 percent on the Multimodal Knowledge Agent Benchmark and 71.6 percent on Agents Last Exam v1.1. For open agent tasks, it ranks fifth in the Agent Arena, first on the Vals Tax Agent Bench with 52.3 percent, and shares first place on the Vals Legal Research Bench with 55.8 percent.
Google is rolling out Hydrogen globally across all platforms. Developers can access it via the Gemini API in AI Studio and Android Studio, while business customers can use Gemini for Business. It is available in the Gemini app and Google Search, and for paying Pro and Ultra subscribers via Google Antigravity. In the Gemini app, the model can include context from Gmail, Drive, Airtable, and Dropbox.
Within the Antigravity coding tool, an orchestrator now dynamically switches between Carbon and Hydrogen. Google claims this orchestrator nearly matches Carbon success rates on Terminal Bench 4.0 while using 24 percent fewer resources and reducing processing time by 19 percent.
An independent evaluation by Artificial Analysis places Hydrogen at the xhigh level roughly equal to Claude Sonnet 5.5. At the high level, its score of 51 is slightly behind GPT 6.1 Sol and Claude Opus 5.5, which both scored 52. Artificial Analysis reports that Hydrogen produces invented statements in only 8 percent of cases on the AA Omniscience benchmark. However, it showed weakness in agent based terminal work on Terminal Bench 4.0 and Terminal Bench Science 0.1, where it scored 54 percent and 50 percent respectively, trailing behind Claude Sonnet 5.5 and GPT 6.1 Sol.