Google brings its new WeatherNext 3 AI forecasting model to Search and Maps

Google DeepMind has released WeatherNext 3, a new AI weather forecasting model, and started rolling it into Google Search, Maps, the Gemini app and Google Cloud. The model produces an updated global forecast every hour instead of every six hours, using live satellite data rather than the delayed output of traditional physics-based simulations.

The jump in detail is large. The earlier WeatherNext 2 model produced forecasts on a 25-kilometer grid in six-hour steps. WeatherNext 3 resolves surface variables like temperature and moisture down to 5 kilometers, other surface data to 10 kilometers, and wind and other atmospheric variables to 25 kilometers, giving a global picture Google says is roughly five times sharper than before.

Better rain and snow forecasts

Precipitation forecasting improved substantially. Google trained the model on NASA’s satellite-based IMERG precipitation data set and its own radar-based precipitation reanalysis. In its own evaluations, Google reports Continuous Ranked Probability Score improvements of up to 60% against IMERG, 30% against the MRMS radar data set, and 10% against rain gauge readings at early lead times. For forecasts a day or more ahead, Google says people will see up to 50% more accurate precipitation predictions, with the largest gains in regions that have historically had less reliable forecasting, including parts of Latin America, Africa and Asia-Pacific.

The model also adds forecasts built for renewable energy planning, including wind speed at roughly turbine height, plus cloud cover and sunlight levels for solar farms, information grid operators can use to match clean energy output with demand. According to Google, independent live evaluations from Brightband currently rank WeatherNext 3 as the most accurate global weather model tested.

The model is live today in Search, the Gemini app, Maps, the Google Maps Platform Weather API and Earth Engine. Developers and researchers can also query the hourly forecast data directly through BigQuery or download it in bulk from Google Cloud Storage.

Google has not published a full technical paper or independent third-party verification beyond Brightband’s live leaderboard, so how the model performs against national weather services over time, particularly in the underserved regions Google highlights, is something to track as more forecasts accumulate.