Google’s newest AI model is its fastest Gemini 4 yet, and on one widely watched leaderboard it edges past the company’s own premium model. Gemini 4 Hydrogen, announced on November 12, 2026, scores 51 on the Artificial Analysis Intelligence Index at its default setting, one point above Gemini 4 Argon, and reaches 54 at its highest thinking level.
It is rolling out globally from today across the Gemini app, Google Search, Google AI Studio, the Gemini API, Gemini for business and, for paid Google AI Pro and Ultra subscribers, the Google Antigravity coding tool.
What Hydrogen is for
Google describes a split in its lineup. Gemini 4 Carbon and Argon handle slow, deep reasoning on hard problems. Hydrogen is meant for quick execution: writing code, running tests, checking changes, and the many small steps inside agent workflows. It takes over the role that the Flash models used to fill, without the Flash name.
Against Gemini 3.8 Flash, released in September, Google says Hydrogen:
- finishes tasks 2.5 times faster at its default “high” setting
- scores 57% on Terminal-Bench 4.1, up from 18.2%
- makes more than five times fewer fabricated claims
- handles failed tool calls and empty search results more gracefully
The model reads up to 1 million tokens of input and can write up to 1 million tokens of output. It offers four thinking levels: low, medium, high (the default) and a new “xhigh” tier.
Price, and what it means in practice
Developers pay a flat rate across the whole context window: $1.50 for every million tokens sent in and $7.50 for every million the model writes back. To put that in plain terms, feeding the model a 200,000-token codebase and getting back a 20,000-token answer would cost about $0.30 plus $0.15, or roughly 45 cents per request.
Changes in the Gemini app and Search
In the Gemini app, Hydrogen appears in the model picker. Google says it sticks more closely to a user’s tone, format and rules, so drafts and study guides need fewer rewrites, and that it calls connected tools such as Gmail, Google Drive, Airtable and Dropbox more reliably. Free users get it only when the app’s automatic model selection decides it suits the prompt.
In Search, Hydrogen powers AI Mode. Google says it handles rambling multi-part questions and typos better, gives more visual answers and widens the range of links it shows by upgrading the “query fan-out” technique behind those answers.
Orchestrator: two models in one session
Alongside the model, Google added Orchestrator to Antigravity. It switches between Carbon and Hydrogen within a single session, so the heavier model plans and the lighter one does routine work. On Terminal-Bench 4.0, Google says this setup came close to Carbon’s success rate while using 24% less quota and finishing 19% faster.
Safety claims
Google reports that Hydrogen cuts overreach in agent tasks by up to 81% versus 3.8 Flash, is up to twice as unlikely to sidestep permission limits on blocked tasks, and scored 96.7% on the company’s internal agent honesty test. These are Google’s own evaluations; outside testing will take longer.
For more on Google’s 2026 AI push, read about the Gemini glasses and their recording light, DeepMind’s AlphaProof Code verifying Signal’s encryption, and Sergey Brin’s comments on Gemini models helping build their successors.
Gemini in Google Sheets gets Hydrogen first for AI Pro and Ultra subscribers, with Google claiming outputs over 30% faster than earlier models in the Sheets canvas.
Image: Runner1928, CC BY-SA 3.0 via Wikimedia Commons
This article was produced with AI assistance and reviewed by our Editorial Desk. How we work.
