Google WeatherNext 3 Brings Hourly, High Resolution AI Forecasts to Search, Gemini and Maps
Google's WeatherNext 3 adds hourly satellite driven forecasts, 5 km detail and sharper rain predictions across Search, Gemini, Maps and Cloud worldwide.
Summary
On September 3, 2026, Google DeepMind and Google Research launched WeatherNext 3, ranked the most accurate global weather model in independent live evaluations by Brightband. It combines live hourly geostationary satellite mosaics, historical analysis and sparse station observations, avoiding the 6 hour lag of supercomputer based numerical weather prediction data used by WeatherNext 2. It refreshes hourly rather than every 6 hours and produces dense grids, cyclone tracks and station level forecasts.
WeatherNext 3 resolves temperature and moisture at 5 kilometers, other surface variables at 10 kilometers and atmospheric variables such as wind at 25 kilometers, making its global picture about five times sharper than WeatherNext 2's 25 kilometer grid. Renewable energy outputs include 100 meter turbine height winds, cloud cover and solar radiation. Training with NASA's Integrated Multi-satellite Retrievals for GPM, or IMERG, and Google's satellite radar precipitation reanalysis improved medium range CRPS by up to 60% against IMERG, 30% against MRMS and 10% against rain gauges at early lead times. Forecasts made at least a day ahead deliver up to 50% more accurate precipitation predictions, with the largest gains in historically less reliable regions.
The model is rolling out globally through Google Search, the Gemini app, Google Maps, Google Maps Platform Weather API and Google Earth Engine. Researchers, developers and businesses can query forecasts through BigQuery and Earth Engine or bulk download them from Google Cloud Storage without model setup. Google targets responders, aviation, farming, supply chains and clean energy, including underserved areas of Latin America, Africa and Asia Pacific. Official warnings and safety advice should still come from meteorological agencies or national weather services.
Positives
- 5 kilometer temperature and moisture forecasts make WeatherNext 3's global picture about five times sharper than WeatherNext 2.
- Hourly satellite observations replace reliance on numerical weather prediction data carrying a 6 hour lag.
- Precipitation CRPS improves by up to 60% against IMERG, 30% against MRMS and 10% against rain gauges.
- 100 meter wind speeds, cloud cover and solar radiation forecasts can improve wind and solar generation planning.
- Google Search, Gemini, Maps, Earth Engine, BigQuery and Cloud Storage provide broad access without model setup.
Risks & concerns
- 25 kilometer resolution still applies to atmospheric variables, compared with 5 kilometers for temperature and moisture.
- 10% improvement against rain gauges is substantially smaller than the 60% gain measured against IMERG.
- Weather remains inherently unpredictable despite hourly updates and higher resolution.
- Google directs users to meteorological agencies or national weather services for official warnings and public safety advice.