Scientists at Google Deepmind and Google Research today released WeatherNext 3, a new artificial intelligence model designed for weather forecasting. Google has indicated that this model will begin feeding into weather information displayed in Google Search, Google Maps, and Gemini, and will also be available on Google's cloud platforms for users and researchers.
According to Google, WeatherNext 3 has demonstrated superior accuracy compared to other leading models when tested on Operational WeatherBench, a tool for comparing AI forecasts. This includes deep-learning models from Google, Microsoft, Nvidia, and the European Center for Medium-Range Weather Forecasting, as well as traditional forecasts from the US National Weather Service and the ECMWF.
The new model addresses previous limitations of AI forecasting, offering predictions down to a 5km resolution on key variables. Its evaluations on rain are reportedly 60% improved over its predecessor, WeatherNext 2, and it can produce hourly forecasts instead of the previous standard of every six hours. These improvements are partly due to WeatherNext 3 being a larger model with 2.4 times more parameters.
WeatherNext 3 is also capable of ingesting real-time hourly weather satellite data. Google states it is the first AI model to directly incorporate raw observations for a high-resolution global forecast, although the AI weather startup WindBorne claims its WeatherMesh 6 model has been doing so since late 2025.