Google launched Gemini 4 Hydrogen on Wednesday, rolling out the new model in the Gemini app and developer channels following this week’s Carbon availability.
The release skips the previously used Flash branding to deliver a major standalone upgrade focused on speed and reasoning capabilities.
“tasks 2.5 times faster at its high default setting” Hydrogen finishes, according to Google, compared to Gemini 3.8 Flash, which launched in September. Google said that the model more than triples Terminal-Bench 4.1 scores, moving from 18.2% to 57%, while cutting fabricated claims by more than five times.
The model offers a native 1 million-token input and up to 1 million output tokens. On benchmarks, Hydrogen surpasses Argon on the Artificial Analysis Intelligence Index at its default high thinking level, scoring 51 compared to Argon’s 50, and reaching 54 at its highest xhigh thinking setting.
In the Gemini app, free users access Hydrogen via automatic model selection when it fits the prompt. The model improves adherence to tone, format, and rules while producing clear tables and visual breakdowns.
Google Search integrates Hydrogen as the default model for AI Mode and uses it for complex questions in AI Overviews. The system processes multi-part queries, typos, and specific requests while generating more visual responses with wider link ranges.
For developers, Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens. In Google Antigravity, the model features an Extra thinking tier alongside Low, Medium, and High settings.
Google introduced Orchestrator in Google Antigravity to let models like Carbon and Hydrogen collaborate within the same session. Paid Google AI Pro and Ultra subscribers can use the feature to let Carbon act as a coordinator while Hydrogen powers worker and verifier agents.
Partners running Hydrogen saw speed increases between 2x and 2.8x alongside token usage reductions of 12% to 23% compared to Gemini 3.8 Flash. The model ranks #5 on Agent Arena for multi-step tasks such as deep research and codebase navigation.
Hydrogen leads its class on HLE-Diamond at 58.8%, ranks #1 overall on Vals Tax Agent Bench at 52.3%, and ties for #1 on the Vals Legal Research Bench at 55.8%. It also leads its class on AutomationBench-AA.
