Google Gemini 4 Hydrogen Launches for Agentic Workflows and Search

Google began rolling out Gemini 4 Hydrogen on November 12, 2026. The model is the third release in the Gemini 4 family and serves as the daily driver option after Carbon and Argon.

Google Gemini 4 Hydrogen focuses on speed and efficiency for coding tasks, multi-step agentic workflows, and routine Search queries. It replaces the earlier Gemini 3.8 Flash model across API, apps, and Search surfaces.

Google Gemini 4 Hydrogen Targets Agentic Workflows With New Specs

The model ships with a 1,048,576-token unified context window that covers input, thinking, and output in one block. It accepts text, images, audio, and video. Four thinking levels are available: low, medium, high default, and xhigh. The model ID is listed as gemini-4-hydrogen.

Benchmarks show clear gains over Gemini 3.8 Flash. At the default high setting the model runs 2.5 times faster. It scores more than three times higher on Terminal-Bench 4.1, moving from 18.2 percent to 57 percent. Fabricated claims drop by a factor of five, and tool-failure recovery improves. On the Artificial Analysis Intelligence Index it reaches 51 at high and 54 at xhigh, above Argon’s 50. It leads or ties for first place on several agentic tests, including Vals Tax Agent and Vals Legal Research.

API pricing stays flat across the full 1-million-token context: 1.50 dollars per million input tokens and 7.50 dollars per million output tokens. The model was adapted from Carbon and adds an orchestrator mode in Antigravity that pairs Hydrogen with Carbon for longer tasks.

Google Gemini 4 Hydrogen Reaches Users Through Search and Enterprise Tools

Availability started globally on November 12. The model appears in Gemini API and Google AI Studio for all tiers, the Gemini app and Google Search for every user, and Antigravity, Sheets, and Chrome for Pro and Ultra subscribers. Business customers and Android Studio users also gain access.

In Google Search the model powers AI Overviews and other features for all users. Google states it improves intent understanding for multi-part and typo-heavy queries, produces more visual responses, and surfaces a wider set of links. Free-tier users can select the model through the Auto option in the Gemini app.

The launch continues a rapid cadence of three Gemini 4 models released in roughly six weeks. Google follows its standard 45-day retirement notice policy for any future changes. Additional benchmark evaluations are expected in the coming weeks as enterprise testing expands.