Google has introduced Gemini 4 Hydrogen, a lightweight artificial intelligence model optimized for fast coding and agent workflows. The release prioritizes low latency and lower operational costs while working alongside larger reasoning models like Carbon and Argon. Access is open globally across Google Search, the Gemini consumer app, and developer platforms starting at $1.50 per million input tokens.
Pricing stays uniform across the entire 1 million token context window at $1.50 for 1 million input tokens and $7.50 for 1 million output tokens. On the Artificial Analysis Intelligence Index, Hydrogen scored 51 points on its default setting and reached 54 points on its highest configuration. Against the older Gemini 3.8 Flash, the new engine completes tasks 2.5 times faster. Test data shows Terminal Bench scores climbed from 18.2% to 57%, while fabricated answers dropped by 5x.
Google splits complex programming workloads between models inside Google Antigravity using an automated tool called Orchestrator. In this setup, the larger Carbon model handles system design while Hydrogen executes code and runs automated tests in parallel workspaces. During benchmark runs on Terminal Bench, this collaborative approach completed tasks 19% faster and consumed 24% less quota than running Carbon on its own. Andrew Carr, cofounder at Cartwheel, evaluated the code checking capabilities during early partner testing:
The new Gemini 4 Hydrogen model is thorough and skeptical in a way I have not seen from past Gemini models.
Google integrated Hydrogen directly into its primary search engine for all global users. The updated algorithm breaks multi part questions with typos into separate queries simultaneously to gather relevant links. Replacing older backends in AI Mode, the system generates visual comparison tables, charts, and interactive cards instead of plain text blocks. Google Vice President Robbie Stein described the rollout as the most substantial search overhaul in the history of the product.
