How does Google’s new weather AI make your forecast more accurate?

WeatherNext 3 combines much finer spatial detail with hourly updates based on recent satellite observations.

WeatherNext 3, announced September 3, starts with fresh global geostationary-satellite imagery and updates hourly instead of waiting for six-hour cycles. It resolves temperature and moisture on a 5 km grid rather than the earlier 25 km grid, which can better catch fast-forming local patterns. Google reports up to a 60% precipitation-score improvement against NASA’s IMERG baseline at early lead times, while Brightband’s Operational WeatherBench lists it as the leading surface-temperature system ahead of ECMWF IFS and the US GFS.

The model now feeds Google Search, Maps, and Gemini. The biggest practical gain may be finer forecasting in regions that cannot afford dense local supercomputing—but it is still a forecast, and Google has not presented the model as open weights.