Google just added a fourth member to its Gemini 4 family, and this one is built for speed. Gemini 4 Hydrogen, announced November 12, is the company's new everyday model — the one that shows up in the Gemini app, in Search, and in developers' toolkits when they need answers fast. The surprise isn't that it's quick. It's that the quick model is now beating bigger, pricier models on the benchmarks that matter.
What Hydrogen actually is
Google describes Hydrogen as its fast execution model. It sits below Carbon (the heavyweight for the most complex work) and Argon (code and analysis) in the Gemini 4 lineup, and replaces the old "Flash" branding entirely. In the Gemini app's model picker, Hydrogen is listed as the choice for "Everyday help" — quick answers still go to 3.5 Flash-Lite, detailed research to 3.1 Pro, and the hardest jobs to Carbon. Free users get Hydrogen automatically when the app's Auto mode decides it's the right model for the prompt.
The speed numbers are hard to ignore
Compared with Gemini 3.8 Flash, which launched in September, Google says Hydrogen finishes tasks 2.5 times faster at its default high thinking setting. Its Terminal-Bench 4.1 score jumps from 18.2% to 57% — more than triple. Partners testing the model report it running 2x to 2.8x faster while using 12% to 23% fewer tokens. It also cuts fabricated claims by more than five times, and Google says it is up to twice less likely to bypass permission boundaries when a task is impossible or blocked.
It punches above its weight on benchmarks
Here's the twist: Hydrogen doesn't just win on speed. On the Artificial Analysis Intelligence Index it scores 51 at its default setting, edging out the bigger Argon at 50, and reaches 54 at its highest thinking level. It ranks #5 on Agent Arena for multi-step agentic work, leads its class on HLE-Diamond (58.8%), and tops the Vals Tax Agent Bench. For developers, the price is flat across its 1-million-token context window: $1.50 per million input tokens and $7.50 per million output tokens.
Where you'll actually meet Hydrogen
Hydrogen is the new default model for AI Mode in Google Search and powers the hardest questions in AI Overviews, with more visual answers and a wider range of links. In the Gemini app it sticks to your tone and format better, so email drafts and study guides come back ready to use on the first try. And in Google Antigravity, a new Orchestrator feature lets Carbon hand routine steps to Hydrogen mid-workflow — Carbon coordinates, Hydrogen does the legwork — so teams ship faster without paying Carbon prices for every step.
Why it matters
The old trade-off in AI was simple: fast models were dumb, smart models were slow and expensive. Hydrogen is Google's attempt to kill that trade-off. Google has had a busy month — from Gemini 4 Carbon to robots on a Mercedes factory floor — and if the model that answers in seconds is also the one topping benchmarks, the "which model should I use?" question starts to disappear — and that is exactly what Google wants as it pushes Gemini into Search, the Gemini app, and developers' daily workflows. The fast model being the smart one changes what everyday AI feels like.
Hydrogen is available now in the Gemini app, Google Search, AI Studio, and the Gemini API. The speed race in AI just got a lot more interesting — because this time, speed didn't cost intelligence.
Comments
Post a Comment