Google's new everyday AI does tasks 2.5x faster than the one it replaces
Google released Gemini 4 Hydrogen, a quicker, lower-cost AI model for daily work, and put it into Search, the Gemini app and its business tools. Teams on Google's older fast model get the biggest jump.
The short version
- Google released Gemini 4 Hydrogen on 12 November. It is the fast, lower-cost member of Google's newest family of AI models.
- Google says it finishes tasks 2.5 times faster than the model it replaces, and makes things up far less often.
- It is live now for developers and businesses. People using Google Search and the Gemini app will meet it too.
What Google released
Google now offers its newest AI in three sizes. Carbon and Argon are the heavy thinkers, built for long, hard problems. Hydrogen is the quick one, made for the many small jobs a team hands to AI every day.
On the Artificial Analysis Intelligence Index, an outside scorecard, Hydrogen scored 51 at its default setting. Argon, its bigger sibling, scored 50. At Hydrogen's highest setting it reached 54.
Developers pay $1.50 for every million chunks of text the model reads, and $7.50 for every million it writes. That price holds however long the document is, up to the model's limit of 1 million chunks.
What is actually new
The jump is clearest for teams on Google's previous fast model. Google says Hydrogen finishes tasks 2.5 times faster at its default setting. It also makes things up less than a fifth as often.
Early business testers saw it run 2 to 2.8 times faster on data-heavy work. It also needed 12% to 23% less text to get there.
It copes with long, mixed files too. It can read PDFs and spreadsheets. It can also work through audio and recorded video, then write a report with its sources marked. In Google Sheets, Google says it builds financial forecast models over 30% faster than before.
Google's coding tool, Antigravity, gets a new feature that switches between models during a job. The stronger model plans the work. Hydrogen does the routine steps. In Google's test, this came close to the top model's success rate. It used 24% less of the user's allowance and finished 19% faster.
Safety claims worth checking
Google says Hydrogen oversteps its instructions up to 81% less often than the older model. When a task is blocked, it is up to half as likely to sneak past its permission limits. Google also says it resists hidden instructions planted in web pages or files more than 2.3 times better.
These are Google's own tests. Treat them as a starting point for your own checks.
What it means for you
Most of the AI work a company does is routine. Someone summarizes a report or drafts a reply. A faster model at a flat price makes that kind of work cheaper to run thousands of times a day.
Your staff will also meet Hydrogen without choosing it. Google is rolling it out worldwide, so people in Indonesia using Google Search and the Gemini app get it too. Free Gemini users get it when the app decides it suits the request.
For businesses that build on Google, the choice of model now matters more than before. A cheap, fast model for most steps, with a stronger one saved for the hard parts, can cut the bill without hurting results.
What to do
- If your developers use Google's older fast model, test Hydrogen on your own most common tasks before switching.
- Keep your premium model for hard planning, and let Hydrogen handle the steps that repeat.
- Check Google's safety figures against your own data before giving it access to live systems.
- Use the flat price to plan the cost of reading long contracts and annual reports.