Google Releases WeatherNext 3 - Hourly Forecasts at 5km for Surface Variables

Google DeepMind and Google Research released the weather forecasting model WeatherNext 3 on September 3, 2026¹.

• Resolution and refresh rate: Key surface variables such as temperature and moisture are handled at 5km, other surface variables at 10km, and atmospheric variables such as wind speed at 25km, with a forecast produced every hour. The previous generation, WeatherNext 2, ran on a 25km grid in 6-hour increments¹
• A change in training data: Most AI weather models, WeatherNext 2 included, were trained on output from numerical weather prediction (NWP) models, which carry a six-hour data lag. WeatherNext 3 learns directly from live geostationary satellite data and from weather station observations¹
• Where it appears: Starting that day it begins powering weather experiences in Google Search, the Gemini app, Google Maps, the Google Maps Platform Weather API and Google Earth Engine. Developers can query the data in BigQuery and Earth Engine, or bulk-download from Cloud Storage¹

Google frames its claim of “the most advanced and accurate global weather model to date” as resting on independent live evaluations by the third party Brightband¹. For planning a day or more ahead, precipitation forecasts are said to be up to 50% more accurate, with the largest gains in regions where forecasts have historically been less reliable¹. For renewable energy, the model adds 100-meter wind speeds — roughly turbine height — along with high-resolution cloud cover and solar radiation¹. For businesses that schedule equipment, deliveries or construction around the weather, that widens the options for pulling this in as external data. The company published its cyclone forecasting paper in Nature in August.

Sources

1. Introducing WeatherNext 3, our most advanced and accurate global weather AI model - Google official blog, credited to The WeatherNext team (September 3, 2026)