DeepMind releases a new weather forecasting model for more accurate predictions (www.engadget.com)

DeepMind has released WeatherNext 2, an upgraded AI weather model that promises higher-resolution, more accurate global forecasts up to two weeks ahead for temperature, pressure and wind, plus hourly predictions and improved tropical storm-track forecasting (hurricane paths reportedly accurate up to three days out versus two days in the prior model). Google says the new system runs about eight times faster than its predecessor, enabling quicker updates that businesses—energy traders, logistics firms, insurers—can use to make more resilient, weather-sensitive decisions. Technically, the jump comes from a new modeling approach detailed in DeepMind’s paper: instead of chaining ML components adapted from image/video generation that require repeated iterative processing, WeatherNext 2 produces forecasts in a single processing step. That reduces compute costs and latency while preserving or improving accuracy, and helps AI-based forecasting outpace many traditional numerical models. Limitations remain—DeepMind acknowledges gaps in training data that make outlier precipitation events (rare heavy rain or snow) harder to predict—but the model’s speed, temporal granularity, and storm-track gains mark a meaningful advance for operational weather forecasting and real-time decision systems.