2026-11-12 · ← News
Gemini 4 Hydrogen offers 1M-token context at $1.50 and makes model routing the product
Google is positioning Gemini 4 Hydrogen as a faster execution model for coding and agents. It offers a 1 million-token context, costs $1.50 per million input tokens and scores 54 on the Artificial Analysis Intelligence Index at xhigh.
The image could not be loaded.
Google has launched Gemini 4 Hydrogen, a faster model for coding, agentic workflows and everyday work. It costs $1.50 per million input tokens and $7.50 per million output tokens, with flat pricing across a context window of up to 1 million tokens.
Hydrogen combines a 1M-token context with four thinking levels
The model is adapted from Gemini 4 Carbon and accepts text, images, audio and video. It offers low, medium, high and xhigh thinking levels. Default high scores 51 on the Artificial Analysis Intelligence Index, while xhigh reaches 54. Artificial Analysis lists the same scores independently, although it does not yet show complete time-per-task data for Hydrogen.
Hydrogen is rolling out globally through the Gemini API and AI Studio, the Gemini app, Search and Google's enterprise products. Some surfaces carry separate conditions: Antigravity requires Google AI Pro or Ultra, while Gemini in Sheets starts with eligible business plans and Pro or Ultra subscribers.
Google is splitting work between expensive planners and cheaper executors
The deployment pattern matters as much as the model. Orchestrator in Google Antigravity switches between Carbon and Hydrogen within one session. On Terminal-Bench 4.0, Google says it reached near-Carbon task success while using 24% less quota and finishing 19% faster.
This is a practical design for agent systems. A more expensive model plans and handles subtle failures, while a cheaper model writes code, runs tests and verifies changes. Savings come from routing and thinking-level selection, not only the token price.
Strong benchmarks arrive before the full safety report
The model card reports 57.0% on Terminal-Bench 4.1 and 71.6% on Agents' Last Exam v1.1. Google also acknowledges hallucinations, occasional slowness or timeouts and higher token use at stronger thinking levels. The full Safety and Security Report is promised for the coming days.
An internal safety evaluation also shows unjustified refusals increasing by 5.57 percentage points against Gemini 3.8 Flash. Google says most regressions were false positives, but production teams should still test their own requests rather than trust an aggregate chart.
Routing accuracy and actual task cost will determine Hydrogen's role
Teams should measure cost per completed task, output-token consumption, latency on their own workloads and the frequency of human intervention. A model with a low list price can still be expensive if xhigh runs long or repeats work.
Hydrogen succeeds if Orchestrator reliably knows when a cheaper executor is enough and when to hand work to Carbon. A benchmark cannot prove that judgment. Bills and completed-task logs can.
Lilith's verdict
Google is staffing an agent shift where Carbon assigns work and Hydrogen runs between the terminal and the test suite. The winner will be the dispatcher that sends the right job to the right worker without returning an overtime bill.
Sources