Google Gemini 4 Hydrogen Reaches Search, App and API

Google's newest AI model isn't a bigger brain. It's a faster pair of hands. Gemini 4 Hydrogen, which Google introduced on Thursday and kept rolling out through Friday, is the company's fast execution model — built for coding loops, tool calls and multi-step agent work — and it's now reaching Search, the Gemini app, Google AI Studio and the Gemini API worldwide.

The positioning is deliberate. Hydrogen joins Carbon and Argon in the Gemini 4 family as the tier meant for everyday and multistep tasks rather than frontier reasoning marathons. Google DeepMind's Koray Kavukcuoglu describes it as built for rapid execution across diverse agentic workflows — the model that's supposed to do the work, quickly, instead of thinking about it at length.

The benchmark Google leads with is a narrow win. On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default high thinking level against Argon's 50, and reaches 54 at its highest xhigh setting — while, Google says, often completing tasks faster than models in its class. Those are the vendor's numbers, published at launch, not an independent replication.

The pricing is the aggressive part. Hydrogen costs a flat $1.50 per million input tokens and $7.50 per million output tokens across its entire one million-token context window — no surcharge as the context grows, with thinking tokens included in the output price. Batch and Flex processing cost half that. For developers, a flat rate across a million tokens removes the spreadsheet exercise of pricing a long-context job.

Access is broad from day one. Developers get Hydrogen through the Gemini API on free and paid tiers, in AI Studio and Android Studio. Consumers meet it inside Google Search, where it's upgrading AI Mode and AI Overviews, and in the Gemini app — though free app users can't pin the model themselves; Google's router hands them Hydrogen when it decides the prompt suits it. Paid AI Pro and Ultra subscribers can select it in Google Antigravity, and it's part of Gemini for business and Gemini in Google Sheets.

Under the hood, the model card says Hydrogen is adapted from Gemini 4 Carbon, with a knowledge cutoff of March 2026. In practical terms it takes over the job Gemini 3.8 Flash has held — Google's everyday coding and agent model — and Google's own migration notes treat it as the model to test next if you're running Flash today. Each request gets a context window of 1,048,576 tokens shared across input, thinking and output, with 65,536 output tokens by default.

There's a deadline hiding in the pricing tables. Flash's standard rates rise to the same $1.50 and $7.50 levels on January 1, 2027, which turns Hydrogen's launch price into the future price of the whole fast tier. Developers who migrate now aren't chasing a discount so much as arriving early at where the pricing was already going.

The rollout's quiet detail is who chooses the model. In Search, Google says Hydrogen is the default in AI Mode and handles the harder questions in AI Overviews, per coverage of the launch — meaning most of the billions of people who meet this model will never learn its name. In the free Gemini app the same logic applies: Hydrogen arrives through automatic model selection, not a menu pick. The launch is less a product unveiling than a silent swap of the engine under products people already use.

What would change the story is independent measurement. Google's index numbers are launch-day claims, and the Artificial Analysis figures it cites — 51 at default, 54 at maximum thinking — will get tested against rivals like Claude and GPT-class models by third parties in the coming weeks. If Hydrogen holds its scores at its price, the pressure lands on every competitor's fast tier. If it doesn't, the flat $1.50 rate becomes the story instead: the price of good-enough intelligence, set by the company that also owns the search box it answers in.

The real test isn't the index score. It's whether a model priced for volume can hold quality across the unglamorous work it's aimed at — repository edits, debugging loops, sub-agents running other agents' errands. Google has put Hydrogen in front of every Search user on earth, which means the evaluation is already running at a scale no benchmark matches. The results will arrive the way they always do with a default model: in whether people stop noticing it's there.