WeatherNext 3 combines live satellite images with global forecasts. It can give detailed information for wind and solar energy teams.
New to this? Read it in simple words
- Google DeepMind and Google Research announced WeatherNext 3, a weather AI, on September 3. It reads new satellite images and makes a new global forecast every hour.
- It can show details as small as five kilometers. Google says the main forecast can look 15 days ahead.
- It can predict wind near wind turbine height and estimate sunlight reaching the ground. This helps energy teams estimate how much wind or solar power may be available.
- Real use will show where it makes mistakes with local or unusual weather. A good global test result may not be good enough for one city.
- AI model
- A computer program trained on lots of data to write, see, speak or make decisions.
A forecast that starts with fresh data
Google DeepMind and Google Research announced WeatherNext 3 on September 3. The model reads new images from weather satellites every hour. This can help it react faster to changing rain, surface temperatures, and storms.
The system gives different levels of detail. Some local surface results are close to five kilometers, while larger weather patterns use wider areas. Smaller areas can show coasts, hills, and valleys more clearly than an older 25-kilometer grid.
Live satellite images support forecasts from 5 to 25 kilometre detail. The longest main forecast covers 15 days.
Why energy teams may care
The model can predict wind near the height of wind turbines. It can also estimate cloud cover and the amount of sunlight reaching the ground. These facts help energy teams estimate how much wind or solar power may be available.
Google says the main forecast can look 15 days ahead and compare 64 possible futures. Hourly updates cover the next 48 hours. Several possible results are more useful than one exact answer because weather always includes uncertainty.
From research to public products
Google says the forecasts will support Search, Gemini, Maps, Earth Engine, and cloud data services. This means many people may use the model without knowing its name. Real use will show where it makes local or unusual weather mistakes.
Developers should test the places and times that matter to them. A good result in a global test may not be good enough for one city or wind farm. WeatherNext 3 could become important public infrastructure, but local checks are still necessary.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1Primary source · September 3, 2026Introducing WeatherNext 3, our most advanced and accurate global weather AI model Google DeepMind and Google Research
- 2
- 3Research · September 3, 2026Google’s latest AI weather model gives you no excuse to forget your umbrella TechCrunch
We checked Google’s launch post and developer guide for the update times, detail levels, and forecast range. We also used independent reporting for outside context. We treat Google’s performance and “first” claims with care when they have not been confirmed by others.