WeatherNext 2 by Google DeepMind

WeatherNext 2 is Google DeepMind's next-generation AI weather forecasting model, delivering predictions up to 8× faster than traditional numerical weather prediction systems. It provides hourly resolution forecasts and generates hundreds of realistic scenarios using a novel Functional Generative Network. This ensemble approach captures uncertainty and extreme events more effectively than deterministic models. The model is designed for a broad range of users, from meteorologists and climate researchers to businesses relying on weather data for logistics, agriculture, and energy planning. It is also integrated into consumer products like Google Search, Gemini, Pixel Weather, and Google Maps, making advanced forecasts accessible to the public. At its core, WeatherNext 2 uses deep learning trained on decades of historical weather data. Its generative ensemble produces probability distributions rather than single-point forecasts, giving users a richer understanding of possible outcomes. The model runs on Google's infrastructure, enabling rapid inference and scalability. What sets WeatherNext 2 apart is its balance of speed, accuracy, and probabilistic output. Unlike traditional models that require supercomputers and hours to run, WeatherNext 2 delivers competitive or superior performance in minutes. It is available via APIs in Google Earth Engine, BigQuery, and Vertex AI, allowing integration into enterprise workflows.