Google's Gemini 4 Hydrogen Edges Past Argon at $1.50 per Million Tokens

Google's new everyday AI model scores 51 on the Artificial Analysis Intelligence Index, one point above Gemini 4 Argon, one of the company's two frontier models. Gemini 4 Hydrogen, announced on November 12, costs developers $1.50 per million input tokens, against Argon's $2, and is rolling out globally through the Gemini API, the Gemini app and Google Search. For the developers who run thousands of agent tasks a day, it means near-frontier results from the model Google built to be fast and cheap.

The Everyday Model That Outscored a Frontier One

Google now sells three Gemini 4 models with clear jobs. Carbon is its strongest reasoning model, and Argon its efficient frontier one; both are built for deep, long reasoning. Hydrogen is the one meant to run all day. Google calls it "our daily driver for agentic workflows," a model built for rapid execution, and it can also work as the fast hands next to Carbon or Argon.

The surprise is how close the fast model got to the slow ones. On the Artificial Analysis Intelligence Index, an independent score that combines a set of hard tests, Hydrogen reaches 51 at its default setting, against 50 for Argon. Hydrogen also brings a new top thinking level, xhigh, above the usual low, medium and high, and there it reaches 54.

“Today, we're introducing Gemini 4 Hydrogen, bringing cutting-edge intelligence to everyone. — Koray Kavukcuoglu, Chief AI Architect, Google”

Speed is the other half of the pitch. Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at its default setting, and that partners testing it saw it run 2 to 2.8 times faster while using 12% to 23% fewer tokens. The clearest gains show up in the work agents do.

57% on Terminal-Bench, and Fewer Shortcuts

Terminal-Bench tests whether a model can get real jobs done at a computer's command line, the text window where developers install software, run scripts and fix what breaks. It is the daily bread of coding agents, and the jump here is the largest in the announcement.

The gains reach outside code. Google reports 52.3% on Vals' Tax Agent benchmark, first overall, and 55.8% on its Legal Research benchmark, tied for first. It also says Hydrogen cuts misaligned actions and task shortcutting by 12 times, and fabricated claims by more than five times compared with 3.8 Flash: the agent that skips a step or invents a result is the one that costs a developer an afternoon.

Google also tested a split setup it calls Orchestrator, which hands each step to Carbon or Hydrogen as needed. On Terminal-Bench 4.0 it reached near-Carbon success rates while using 24% less quota and finishing 19% faster. That mix only pays off if the fast model is cheap, which brings us to the price.

The Price Math, and Who Gets It Today

Developers pay $1.50 per million input tokens and $7.50 per million output tokens, the same price across Hydrogen's full context window of one million tokens, and it can write up to one million tokens in a single answer. Batch and Flex requests cost half. There is also a free tier in Google AI Studio, where free usage can be used to improve Google's products.

The gap with the frontier models is about to widen. Argon and Carbon cost $2 and $10 until November 30, and $4 and $20 from December 1. Hydrogen does cost twice today's price for Gemini 3.8 Flash, $0.75 and $3.75, though Flash moves to the same $1.50 and $7.50 on January 1, 2027.

It is rolling out everywhere at once: to developers through the Gemini API, AI Studio and Android Studio under the name gemini-4-hydrogen; to Gemini app users in the model menu, and to free users when automatic selection picks it; to everyone in Google Search; and to Google AI Pro and Ultra subscribers in Antigravity and in Gemini in Google Sheets.

The model Google built to be the cheap one now scores above one of its frontier models, and it is already answering questions in Search.