Google says Hydrogen completes tasks 2.5 times faster than Gemini 3.8 Flash, even when operating at its high reasoning setting. It also reportedly reduces fabricated claims by more than fivefold and delivers substantial improvements in coding performance, including more than three times the previous model's score on Terminal-Bench.
Hydrogen offers four reasoning settings: low, medium, high, and a new extra-high option called xhigh. Its performance also compares favorably with Google's more powerful models in certain evaluations. Hydrogen scored 54 on the Artificial Analysis Intelligence Index at xhigh, compared with 50 for Gemini 4 Argon. Google also highlighted Hydrogen's fifth-place ranking on Agent Arena benchmarks covering multi-step activities such as research, data analysis, and navigating large codebases.
One of Hydrogen's most significant applications is Google Search, where the model is being introduced to users worldwide. Google says the upgrade improves Search's ability to interpret complicated requests, including questions containing multiple parts, specific instructions, or spelling errors.
The improvements extend to AI Overviews and AI Mode, where users can expect more detailed and visually organized answers. Google is introducing additional images, comparison tables, structured information, and customized tools intended to make search results easier to navigate.
Hydrogen also strengthens Google's query fan-out technology, which divides complex questions into multiple related searches. According to Google, the updated system can explore relevant subtopics more effectively and surface a wider selection of websites. The company has previously promised improvements to how its AI-powered search features connect users with external sources.
Multilingual performance is another focus. Google says Hydrogen provides more precise responses across all languages supported by AI Mode, improving its ability to interpret detailed questions and follow instructions.
Beyond Search, Google is positioning Hydrogen as an efficient option for developers building AI-powered applications and automated coding workflows. The model can independently generate code, execute tests, and verify changes, while also working alongside Carbon and Argon when more advanced reasoning is required.
Google continues to recommend its more powerful models for demanding, long-running tasks and security-related work. Hydrogen instead offers a faster alternative for routine development activities, helping reduce token consumption while maintaining strong performance.
Developers can access Hydrogen through the Gemini API at $1.50 per million input tokens and $7.50 per million output tokens, matching the pricing of Gemini 3.8 Flash. By comparison, Carbon and Argon currently cost $2 per million input tokens and $10 per million output tokens, with prices scheduled to increase to $4 and $20, respectively, next month.
While Hydrogen is available to all Google Search users globally, access varies across Google's other products. The model is available to AI Pro and Ultra subscribers in the Gemini app, while AI Plus and free users remain on Gemini 3.5 Flash-Lite.
Google is also bringing Hydrogen to AI Studio, the Gemini API, and its Antigravity development platform, including access for paid AI Pro and Ultra subscribers. The company has indicated broader availability across AI Studio tiers, although access conditions differ by product.
With Hydrogen, Google is expanding its Gemini 4 lineup beyond its highest-performing reasoning models. The release emphasizes faster execution, lower operating costs, and wider deployment, particularly in Search, where the upgraded model is being made available to users worldwide.
This analysis is based on reporting from Google & ars TECHNICA.
Image courtesy of Google.
This article was generated with AI assistance and reviewed for accuracy and quality.
