Google's AI Weather Model Beats Government Forecasts

Google DeepMind and Google Research put out a new weather forecasting model on Thursday, and the company says it's already the most accurate system tested on Operational WeatherBench, a benchmark built by the startup Brightband that tracks metrics like temperature, windspeed, and humidity. The model, called WeatherNext 3, is set to feed into Search, Google Maps, and Gemini, marking one of the first times an AI weather system will power core features across Google's consumer products rather than staying confined to research papers.

The release lands at a moment when deep learning has upended a field long dominated by government supercomputers. Agencies like the US National Weather Service and the European Center for Medium-Range Weather Forecasting have spent decades refining physics-based models that churn through equations describing atmospheric behavior. Those systems work, but they're slow and costly to run.

That started to change in 2018, when the ECMWF released over 50 years of weather data, giving researchers the raw material to train neural networks that could match traditional forecasts at a fraction of the computing time. WeatherNext 3 is Google's latest entry in that race, and according to the company's own testing, it now edges out rivals built by Microsoft, Nvidia, and the ECMWF itself.