Google spent two days this week handing its strongest Gemini models to developers and businesses. Today it shipped the one most people will actually touch.
Google introduced Gemini 4 Hydrogen on Thursday, November 12, calling it the fast, efficient member of the Gemini 4 family, built for everyday coding and agentic workflows instead of the deep, long horizon reasoning that Gemini 4 Carbon and Gemini 4 Argon are for. The rollout is the broad one: all tiers of Google AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search for all users, and Google Antigravity for paid Google AI Pro and Ultra subscribers. Google’s own line for the launch is that Hydrogen brings Gemini 4 to everyone, and for once the distribution list backs the slogan up.
Where Hydrogen sits in the family
The Gemini 4 lineup now has a clear division of labor, at least in Google’s framing. Carbon and Argon sit at the top and take the hardest jobs, the long reasoning runs where a model spends more tokens and checks its own work. Hydrogen is the execution engine underneath them: writing code, running tests, verifying changes, grinding through multi step tasks quickly and cheaply enough that you do not think twice about letting it run.
Google is also using Hydrogen to sell a pairing. A new feature called Orchestrator in Google Antigravity switches between models inside one session, so Carbon can handle the architecture calls while Hydrogen does the fast building and testing in parallel workspaces. Google says Orchestrator reached near Carbon task success rates on Terminal-Bench 4.0 while using 24 percent less quota and finishing 19 percent faster on average. That is Google grading its own homework, so file it as a claim, but the idea itself is the interesting part: one expensive brain, one cheap pair of hands, one bill.
The price, and why the flat part matters
Hydrogen costs 1.50 dollars per million input tokens and 7.50 dollars per million output tokens, and Google stresses that the price is flat across the entire 1 million token context window. Plenty of models charge more once a conversation or a codebase gets long. This one does not, at least at launch.
Stack it against the week. Gemini 4 Carbon, which went generally available on Tuesday, costs 2 dollars in and 10 dollars out at its introductory price. Claude Fable 5.2, launched Monday, costs 8 dollars in and 40 dollars out. Hydrogen undercuts Carbon by a quarter on both sides and lands at less than a fifth of Fable’s output price. It is not trying to beat those two at their own game, and Google does not pretend otherwise. It is priced so that the everyday work, which is most of the work, never needs to visit the expensive tier at all.
The claims, labeled as claims
Everything in this section is Google reporting its own results.
Google says Hydrogen beats Argon on the Artificial Analysis Intelligence Index at its default high thinking level, 51 to 50, and reaches 54 at a new xhigh thinking setting, while often finishing faster than models in its class. Against its own predecessor, Gemini 3.8 Flash, Google says Hydrogen finishes tasks 2.5 times faster at the high setting, more than triples its Terminal-Bench 4.1 score (57 percent against 18.2 percent), and cuts fabricated claims by more than 5 times. It carries a 1 million token input window and can output up to 1 million tokens on multimodal knowledge work, and Google lists class leading or top results on Agent Arena (number 5 overall), HLE-Diamond (58.8 percent), the Vals Tax Agent Bench (number 1 at 52.3 percent) and a new Multimodal Knowledge Agent benchmark (53.4 percent). On safety, Google claims agent overreach down by up to 81 percent and misaligned actions cut by 12 times against 3.8 Flash, with a 96.7 percent score on its internal agent honesty evaluation.
The pattern to notice is that almost every number compares Hydrogen to Google’s own older model or sits on a benchmark Google helped define. Independent testing will sort out how much of it survives contact with other people’s workloads. The price and the rollout, though, need no benchmark to verify: both are live today.
What changes if you just use the app
You do not need to be a developer to meet this model. Hydrogen is rolling into the Gemini app model picker and into Google Search for everyone, and Google says Search answers get better at reading messy, multi part questions and at returning visual, scannable results with a wider range of links. In the app, the pitch is follow through: sticking to your format and tone, calling the right connected tools, and finishing multi step tasks with less back and forth. Four thinking levels (low, medium, high by default, and the new xhigh) let the same model trade speed against depth per task.
My take
Monday was Anthropic cutting prices. Tuesday was Google opening Carbon to everyone at a quarter of Claude’s price. Thursday is Google going one rung lower and saying the fast model is the product now. I read that sequence as the frontier labs quietly agreeing on something: the race is no longer only about the smartest model in the room. It is about which company owns the cheap, fast, good enough layer where a billion everyday tasks actually happen, because that layer is where the habit forms and the subscriptions stick.
For readers of the comparison piece on this site, the practical change is simple. Gemini’s free and everyday experience just got a real model behind it, not a cut down leftover, and anyone paying API bills gets a flat rate they can actually predict. Whether Hydrogen is as sharp as Google’s charts say is a question for independent tests. That it is here, everywhere, today, at 1.50 and 7.50, is not a claim. It is a launch.
Sources
Google, “Introducing Gemini 4 Hydrogen for Agentic Workflows”, official Google blog, November 12, 2026
Google, “Gemini 4 Carbon is now available”, official Google blog, November 10, 2026 (published on this site, post 2350)
Anthropic, Claude Fable 5.2 and Mythos 5.2 launch, November 9, 2026 (published on this site, post 2316)
The Checkpoint Poll
Leave a Reply