Google launches WeatherNext 3 AI model with live satellite data for faster, more detailed forecasts

According to Google, WeatherNext 3 can generate forecasts each hour and visualize conditions such as temperature and moisture at up to a 5-kilometer resolution, compared with the previous WeatherNext 2 model, which produced forecasts every six hours on a 25-kilometer grid. The company calls it its most advanced and accurate global weather model yet.

The model is trained to use real-time satellite observations rather than relying only on output from traditional numerical weather prediction models, which typically involve a six-hour lag. Google says this gives the model a continuously updating view of the atmosphere and helps fill gaps in regions with sparse ground-based rain gauges, particularly outside the United States and Europe. It also combines precipitation data from NASA with Google's own satellite analysis.

For precipitation forecasts made a day or more in advance, Google says WeatherNext 3 can be up to 50 percent more accurate than earlier versions, with the largest gains in areas where forecasts have historically been less reliable. TechCrunch, however, reported that the model's evaluations on rain are 60 percent improved over WeatherNext 2, citing Google researchers. The Verge and Engadget both cited Google's 50 percent figure. WeatherNext 3's capabilities have also been measured on Operational WeatherBench, where, according to TechCrunch, it beat other deep-learning models from Google, Microsoft, Nvidia and the European Centre for Medium-Range Weather Forecasts, as well as traditional forecasts from the U.S. National Weather Service and ECMWF.

The model has 2.4 times more parameters than its predecessor and is designed to target forecasts at specific weather data stations, which Google says improves granularity and allows evaluation against ground-truth measurements. Samier Merchant, a senior staff engineer at Google Research, told TechCrunch that core variables from the model will power many Google products. WeatherNext 3 is now incorporated into Google Search, Maps, Gemini, the Google Maps Weather API and Google Earth Engine. It is also aimed at helping renewable-energy operators by predicting 100-meter wind speeds near turbine height, along with cloud cover and solar radiation, according to Engadget. Ferran Alet, a research scientist at Google DeepMind, told The Verge that as energy needs grow, making renewable energy an appealing option is important.

The launch is the latest step in the growing use of deep learning in meteorology. While traditional forecasts from supercomputers simulate atmospheric physics, AI models learn patterns from historical data and can make faster predictions. Google says WeatherNext 3 is the first AI model to directly incorporate raw observations for a high-resolution global forecast, but WindBorne, an AI weather startup, told TechCrunch that its WeatherMesh 6 model has ingested raw observations from its weather balloons and other sources since late 2025. Google responded that its forecasts have higher resolution globally. Both models still rely on national weather datasets for many inputs, according to TechCrunch.

The original WeatherNext model has been open-sourced since August 2026, and WeatherNext 3 is available to try in Google Weather Lab. Google has also worked with the U.S. National Hurricane Center and agencies in Asia to improve weather forecasting with its AI models.