Google's Gemini 4 Hydrogen: $1.50/$7.50, in Search and Sheets today

Google released Gemini 4 Hydrogen on 12 November 2026, the cheaper, faster member of its Gemini 4 family, and switched it on the same day in the products office teams already use: the model menu in Gemini for business, the Gemini app, Gemini in Google Sheets "for eligible business subscriptions", and Google Search for all users. Google's launch post prices it at $1.50 per million input tokens and $7.50 per million output tokens (tokens are the chunks of text a model reads and writes, and what usage-based AI bills count).

For an IT lead who administers Gemini for business, a new option is in the model menu, with no seat-price change announced. For anyone paying usage on Google's larger models, directly or through a vendor, the 1 December price rise now has a cheaper way round it. For marketing, the model that writes Google's AI answers changed today.

What it costs against Argon and Carbon

Gemini 4 Argon and Carbon are both $2 input and $10 output per million tokens until 30 November, then $4 and $20, per Google's docs as we reported on 13 October and 10 November. Hydrogen is 25% below them today ($1.50 ÷ $2) and 62.5% below from 1 December ($1.50 ÷ $4 = 37.5% of the price; the output rate works out the same). A job that reads 1 million tokens and writes 200,000 costs $1.50 + $1.50 = $3.00 on Hydrogen, against $4.00 on Argon today and $8.00 from 1 December. The job mix is ours, to show the sum.

The saving only counts if the cheaper model does the job. Google says Hydrogen scores 51 on the Artificial Analysis Intelligence Index against Argon's 50, and finishes tasks 2.5 times faster than Gemini 3.8 Flash, the model Google pitches it as the upgrade from, while cutting fabricated claims "by more than 5x". Google names no customer and gives no error rate on real office work.

What changes in the tools

• Gemini for business: Hydrogen is available "via the API or model drop down". Google does not say whether it becomes the default or whether an admin can set one.
• Sheets: Google says Hydrogen turns financial data into scenario models and produces outputs "over 30% faster than previous models". It gives no accuracy figure for the formulas.
• Tax and legal work: Google cites a first place on Vals AI's Tax Agent Bench at 52.3% and a tie for first on its Legal Research Bench at 55.8%. The best score on each leaves close to half the marks unearned, so a person still checks the output.
• Search: Google calls this "the biggest model upgrade to Search yet" and says answers will run more simultaneous searches, link to "a broader range" of sites and show more tables and images. It publishes no figure on clicks sent to websites.
• Safety: Google says Hydrogen is more than 2.3 times more resistant to indirect prompt injection (instructions hidden in a web page or document that try to hijack the assistant) on the Gray Swan benchmark. The figure is Google's.

The pushback

On Hacker News, one commenter argues Hydrogen costs the same as 3.8 Flash's undiscounted rate and more than Flash customers pay today, and asks whether Google is dropping its cheap tier. We could not check 3.8 Flash's list price this run, and Google's post gives no retirement date for it. Another commenter reports early Hydrogen runs that did more than was asked; Google claims up to 81% less of that overreach than 3.8 Flash. Neither measures your workload. The post says nothing about training for staff or who reviews output.

What to do

• IT: ask every vendor whose product runs on Gemini, and whoever owns in-house automations, which model each job uses and what the bill becomes on 1 December. Move a job to Hydrogen only after re-running 50 to 100 real cases and comparing the answers.
• IT: tell Gemini for business users which model to pick for which work, since Google has not said what the default is.
• Marketing: record this week's Google traffic and AI-answer citations for your top pages as a baseline, then compare in two weeks.