Google DeepMind Launches Gemini 4 Hydrogen for Efficient Coding and Agentic Workflows

Google DeepMind released Gemini 4 Hydrogen, a model optimized for speed and efficiency in everyday coding and agentic workflows, distinct from the long-reasoning frontier models Gemini 4 Carbon and Gemini 4 Argon.
Performance benchmarks show a 51 score on the Artificial Analysis Intelligence Index at the default high thinking level (54 at xhigh), 57% on Terminal-Bench 4.1 (over 3x higher than Gemini 3.8 Flash's 18.2%), and 58.8% on HLE-Diamond, where it leads its class.
Compared to Gemini 3.8 Flash, Hydrogen delivers 2.5x faster task completion at the high thinking level, reduces fabricated claims by over 5x, and cuts token usage by 12-23% in data-centric workflows while maintaining accuracy.
The model introduces a new xhigh thinking level alongside low, medium, and high, allowing predictable scaling of quality, latency, and cost, with a fixed price of $1.50 per million input tokens and $7.50 per million output tokens across a 1 million token context window.

Availability and Integration

Available globally via Google AI Studio, Gemini API, Gemini for business, the Gemini app, Google Search, and Google Antigravity for AI Pro and Ultra subscribers.
Google Antigravity introduces an Orchestrator that dynamically switches between models like Carbon and Hydrogen, reducing quota usage by 24% and improving processing speed by 19% on Terminal-Bench 4.0 tasks.
Free Gemini app users can access Hydrogen when the system automatically selects it for appropriate prompts, though most queries default to Flash-Lite or Flash.

Safety metrics show up to 81% reduction in overreach and 79% reduction in under-service compared to Gemini 3.8 Flash, with a 96.7% integrity rate in internal agent evaluations.
The model maintains 99.9% chain-of-thought readability across over 200,000 audit turns and demonstrates 2.3x higher resilience against indirect prompt injection attacks on the Gray Swan IPI benchmark.