Google Ships Gemini 4 Hydrogen at $1.50 per Million Tokens: the Everyday Gemini 4 Model That Outscores Argon at Default Settings

Google has released Gemini 4 Hydrogen, the third new model in its Gemini 4 family in roughly six weeks, and the first one aimed squarely at everyday work rather than frontier reasoning. The company is positioning Hydrogen as the fast-execution tier of the lineup: where Gemini 4 Carbon and Argon handle long-horizon planning, difficult software engineering and deep research, Hydrogen is built for quick coding, tool calls and the multi-step tasks that agents run all day.

The model is also the first Gemini 4 release that every user can touch. It is rolling out globally across the Gemini API, AI Studio, the Gemini app, Google Search and enterprise offerings, and free users get it through the automatic model selection in the Gemini app. Paid AI Pro and Ultra subscribers can also run it inside Google Antigravity, the company's agentic development environment.

On benchmarks, Google says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high thinking level, one point above the Gemini 4 Argon frontier model, and 54 at the maximum xhigh setting. The flagship Carbon still leads the family at 60. Compared with the previous mid-tier model, Gemini 3.8 Flash, Google reports that Hydrogen completes tasks 2.5 times faster at default settings, lifts Terminal-Bench 4.1 scores from 18.2 percent to 57 percent, and cuts fabricated claims by more than 5 times. All benchmark figures are company-reported and have not been independently verified.

Pricing is flat across the model's 1 million-token context window: $1.50 per million input tokens and $7.50 per million output tokens, with four thinking levels (low, medium, high and xhigh) that trade latency and cost against answer quality. In Search, Hydrogen becomes the default model for AI Mode and takes over the hardest queries in AI Overviews, where Google says it handles multi-part questions, typos and explicit formatting instructions noticeably better than its predecessor.

Google also shipped Orchestrator, a new Antigravity feature that dynamically switches between Carbon and Hydrogen inside a single session depending on task complexity. In internal tests, the company says Orchestrator approaches Carbon's success rate on Terminal-Bench 4.0 while consuming 24 percent less quota and finishing 19 percent faster, with Carbon acting as coordinator and Hydrogen driving parallel build-and-verify agents in isolated workspaces. On safety, Google claims Hydrogen reduces agent out-of-bounds behavior by up to 81 percent and under-serving by up to 79 percent, with 99.9 percent chain-of-thought readability — again, company-reported numbers.

The release completes Google's rapid re-stratification of its model lineup in about six weeks: Carbon at the top for reasoning, Argon in the middle, and Hydrogen as the cheap, fast default. The strategy pushes the "which model should I call" decision into automatic routing, and it puts direct price pressure on rival small models, since a $1.50-per-million-token model now sits inside a free product used by billions. The open question is whether automatic model selection holds up once users realize that the router, not the model, now decides how smart their answer is.