Google rolls out WeatherNext 2 for faster forecasts
Google DeepMind has released WeatherNext 2, an artificial intelligence system that produces weather forecasts eight times faster than its predecessor while generating hundreds of potential scenarios from one data input in less than 60 seconds. The advanced model operates on a single tensor processing unit and outperforms the earlier version on 99.9 percent of measured variables across temperature, wind, and humidity for prediction periods spanning zero to 15 days.
The technology uses a Functional Generative Network that introduces controlled noise into its architecture, enabling physically realistic forecasts across interconnected weather systems. Google has integrated WeatherNext 2 into Search, Maps, Gemini assistant, and Pixel Weather applications, with forecast data accessible through Google Earth Engine and BigQuery for developers and researchers.
The company has launched an early access program on Vertex AI for custom inference capabilities, positioning the system as infrastructure for climate modeling applications ranging from disaster preparedness to agricultural planning and energy grid optimization.
