Gemini 4 Hydrogen and Orchestrator announced
blog.google

Gemini 4 Hydrogen and Orchestrator announced

Archived — this story has rotated out of today’s deck. It is kept here in full.

The gist

Google released Gemini 4 Hydrogen, a 1 million-token model priced at $1.50 per million input tokens.
It skipped Argon's vetting stage and reaches free app users at launch.

Background

Google is the company behind the Gemini family of AI models; Hydrogen is the third Gemini 4 model to ship in just over six weeks, joining Carbon and Argon, which Google positions for more demanding reasoning work. Hydrogen emphasizes rapid execution across coding, research, data analysis and agentic workflows, and it skipped the Fairwind Program vetting stage that Argon went through, so anyone can use it from launch day. Google DeepMind's Koray Kavukcuoglu said in late September that the company wanted Gemini 4 out "much earlier" than the end of the year.

How it unfolded

  1. Sep 2, 2026Gemini 3.8 Flash launched, the predecessor Hydrogen is measured against.
  2. Oct 13, 2026Gemini 4 Argon became available to paying developers after nearly two weeks with members of Google's Fairwind Program.
  3. Nov 10, 2026Gemini 4 Carbon availability opened to paying developers and Google AI Ultra subscribers.
  4. Nov 12, 2026Google announced and released Gemini 4 Hydrogen with a 1 million-token context window and four thinking levels, and added Orchestrator to Google Antigravity, which switches between models such as Carbon and Hydrogen within a session.
  5. Nov 12, 2026Hydrogen rolled out to the Gemini API, Google AI Studio, Gemini for business, the Gemini app, Google Search, and paid Google AI Pro and Ultra subscribers in Antigravity; free app users get it via automatic model selection.
  6. Nov 30, 2026Argon's introductory pricing ends, moving from $2 input / $10 output to $4 and $20 per million tokens.

Key terms

agentic
AI programs that can make decisions, pursue goals, and take actions independently, often used for coding tasks.
context window
The maximum number of tokens a large language model can consider at one time when generating an output.
tokens
A machine learning model trained on huge amounts of text to generate, summarize, translate, and analyze language.

Source: Wikipedia

Who’s saying what

Official
Google says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high setting versus 50 for Argon (54 at xhigh), runs about 2.5 times faster at default high, lifts its Terminal-Bench 4.1 score from 18.2% to 57% versus Gemini 3.8 Flash, and cuts fabricated claims more than fivefold.
Official
Google says Orchestrator approached Carbon's task success rates in Terminal-Bench 4.0 testing while using an average of 24% less quota and finishing tasks 19% faster.
Expert
Andrew Carr, chief scientist at Cartwheel Inc. and a former OpenAI scientist, said the model caught two problems with an approach his team had been stuck on for 10 days.
Caution
CASRAI says the 24% quota and 19% speed figures describe Google's own evaluation rather than measured savings for university software teams.

Sources

See today’s stories in the app Gistline — free on the App Store