Google has announced the release of WeatherNext 3, an updated iteration of its machine learning-based weather forecasting model. While previous versions relied exclusively on atmospheric reanalyses—global snapshots that aggregate diverse data into a consistent model every six hours—WeatherNext 3 incorporates direct satellite weather data. This integration allows the model to bypass some of the information loss inherent in the reanalysis process and significantly reduces the latency between current atmospheric observations and forecast generation. By utilizing raw satellite inputs alongside reanalysis data, Google has successfully increased forecast frequency from six-hour intervals to hourly updates. Furthermore, the update features enhanced spatial resolution and a larger machine-learning architecture, which necessitated internal process optimizations to manage the increased computational load. To improve accuracy regarding rainfall, the system now includes a dedicated secondary machine-learning model specifically trained on satellite-derived precipitation estimates, providing users with multiple precipitation forecast options. This approach maintains the primary advantage of AI-driven forecasting—achieving performance comparable to traditional physics-based models while requiring significantly less computational power—while narrowing the data gap between traditional meteorological methods and pure machine learning approaches.