Google DeepMind and Google Research have unveiled a new artificial intelligence model called WeatherNext 3, designed to improve the accuracy and detail of weather forecasting. This model is expected to gradually be added to several Google services, including Google Search, Google Maps, and Gemini, to enhance their weather-related features. WeatherNext 3 offers a significant improvement in both spatial and temporal resolution compared to its predecessor, WeatherNext 2. While WeatherNext 2 provided forecasts with a resolution of 25 kilometers and updates every six hours, WeatherNext 3 can predict weather conditions at a resolution of 5 kilometers and update forecasts every hour. Testing using the Operational WeatherBench comparison tool shows that WeatherNext 3 performs better than other deep learning models developed by companies like Microsoft and Nvidia, as well as traditional forecasting models from the European Centre for Medium-Range Weather Forecasts (ECMWF) and the United States National Weather Service. To achieve this improved performance, Google has increased the size of the model, giving it 2.4 times more parameters than WeatherNext 2. This increase allows the model to process more complex data and generate more accurate predictions. The training of WeatherNext 3 relies on data from specific meteorological stations, which helps the model create more detailed local forecasts. Additionally, the model uses real-time satellite data that is integrated hourly, allowing for more frequent updates. Unlike earlier models that primarily used processed meteorological data, WeatherNext 3 uses raw observational data directly. This approach enables the model to provide more accurate and timely forecasts. AI-based weather forecasting, such as that used in WeatherNext 3, is faster and more cost-effective than traditional numerical methods, which require powerful supercomputers to solve complex atmospheric equations. Instead of relying on these supercomputers, AI models like WeatherNext 3 learn from historical weather data patterns, reducing the computational resources needed for forecasting. Ferran Alet, a researcher at Google DeepMind, noted that machine learning helps overcome the challenge of predicting complex physical phenomena when information is incomplete and computational resources are limited. By analyzing large volumes of data, the model can identify and predict recurring weather patterns with greater accuracy.