Google DeepMind rolls out WeatherNext 3, an AI weather model powered by live satellite data

Google DeepMind has released WeatherNext 3, a new version of its AI-powered weather forecasting model, which the company says produces more detailed forecasts for variables such as temperature, moisture and wind speed.

A shift to live data

The previous model relied heavily on numerical weather prediction data, which carries roughly a six-hour lag. WeatherNext 3 instead incorporates live satellite data, giving it a continuously updating view of the atmosphere. The model also draws on sparse weather station readings, improving forecasts for variables like humidity, which can vary sharply within just a kilometer.

Higher resolution, better accuracy

The new model can generate hourly forecasts at a higher resolution — a 5-kilometer square grid, compared with the 25-kilometer grid used by WeatherNext 2. By combining precipitation data from NASA with Google's own satellite analysis, DeepMind says WeatherNext 3 delivers up to 50% more accurate precipitation forecasts, with the biggest gains in regions where forecasting has historically been less reliable.

Uses for renewable energy

DeepMind says the model should be especially useful for renewable energy companies, since it can forecast wind speeds at 100 meters — roughly turbine height — for more precise wind-energy output estimates, along with high-resolution cloud cover and solar radiation data to assist solar farms.

Where it will be available

WeatherNext 3 will power weather features in Google Search, the Gemini app and Google Maps, as well as the Google Maps Weather API and Google Earth Engine. Anyone interested can try the model through Google Weather Lab. Developers, meanwhile, can build on the earlier WeatherNext model, which was open-sourced in August 2026.