Google has released WeatherNext 3, a new AI weather model that improves forecast resolution, rainfall accuracy, and update frequency while beginning to feed weather information into Search, Google Maps, and Gemini. The model was developed by Google DeepMind and Google Research and will also be available through Google Cloud for researchers and other users.
Google says WeatherNext 3 can forecast key weather variables at resolutions down to 5 kilometers, compared with the 15-to-25-kilometer areas often used by earlier AI systems. It can also generate hourly forecasts instead of the six-hour intervals common in many weather models.
WeatherNext 3 Improves Rain and Local Forecasting
Google says rainfall evaluations improved by 60% compared with WeatherNext 2. The new model is also trained to forecast conditions at specific weather stations, allowing researchers to compare predictions directly with local observations.
WeatherNext 3 has 2.4 times more parameters than its predecessor and uses redesigned decoder targets to generate more useful outputs. Google has also expanded the model beyond standard gridded forecasts, building on earlier work that allowed its systems to visualize cyclone paths.
The model can ingest hourly satellite observations collected in near real time. Google says this makes WeatherNext 3 its first AI model to directly incorporate raw observations into a high-resolution global forecast.
Weather startup WindBorne has said its WeatherMesh 6 model has also incorporated raw observations since late 2025, including measurements collected by its weather balloons. Google said WeatherNext 3 differs by delivering higher-resolution forecasts globally.
Both systems still depend partly on weather datasets produced by national agencies, meaning neither yet operates entirely from raw observations.
Google Says WeatherNext 3 Leads Current Benchmarks
WeatherNext 3 currently ranks first on Operational WeatherBench, a benchmarking system developed by Brightband that compares forecasts across variables including temperature, wind speed, and humidity.
According to the benchmark, WeatherNext 3 outperformed other AI forecasting systems from Google, Microsoft, Nvidia, and the European Centre for Medium-Range Weather Forecasts. It also scored ahead of traditional forecasts produced by the U.S. National Weather Service and ECMWF on the evaluated measures.
Traditional weather forecasting relies heavily on supercomputers solving equations that represent atmospheric physics. Deep-learning systems instead learn patterns from large historical weather datasets and can generate forecasts much faster and at lower computational cost.
Google says the added resolution could support applications including renewable energy planning, where more precise wind, rain, and cloud forecasts can help operators estimate generation conditions. The company also plans to use WeatherNext 3 across consumer products as its weather data begins powering Search, Maps, and Gemini.
Featured image credits: Google Blog
For more stories like it, click the +Follow button at the top of this page to follow us.