Google DeepMind Launches Gemini 4 Hydrogen to Power High-Speed AI Agents

Google DeepMind has announced the release of Gemini 4 Hydrogen, a new AI model designed to provide high-speed intelligence for developer workflows and autonomous agents. Positioned as a versatile middle-tier option within the Gemini 4 ecosystem, Hydrogen aims to balance the deep reasoning capabilities of its Carbon and Argon counterparts with the rapid execution required for everyday coding and testing tasks.

The model introduces a flexible "thinking level" system, allowing users to choose between low, medium, high, and xhigh settings. At its default high level, Hydrogen achieves a score of 51 on the Artificial Analysis Intelligence Index, surpassing the frontier Argon model in specific benchmarks. Google reports the model operates 2.5 times faster than the previous Gemini 3.8 Flash and features a 1-million-token context window with flat pricing of $1.50 per million input tokens and $7.50 per million output tokens.

Optimized Agentic Workflows

To facilitate complex projects, Google is also launching Orchestrator within its Antigravity platform. This tool automatically transitions between different Gemini 4 models based on the task at hand. According to internal data from Terminal-Bench 4.0, using Orchestrator can reduce quota consumption by 24% and improve completion speeds by 19% compared to using a single high-end model for an entire session.

Hydrogen’s capabilities extend to specialized creative and technical tasks. It powers "Geminisland," a tool that generates 3D voxel environments from single prompts, and can autonomously design complex web interfaces featuring WebGL2 shaders. In productivity suites, the model can transform financial data in Google Sheets into interactive models 30% faster than earlier versions and perform automated layout audits in Google Slides.

What to watch next: The impact of tiered reasoning levels on the broader LLM market as competitors look to match Google's granular pricing and performance options.