Google DeepMind announces WeatherNext 3 forecasting model
Google DeepMind says WeatherNext 3, built with Google Research, learns directly from real-time observations to produce faster, more localized forecasts.
Google DeepMind has announced WeatherNext 3, its new model for global weather forecasting. According to the company's post on X, the model was developed together with Google Research.
The stated difference from earlier work is the training signal: WeatherNext 3 learns directly from real-world, real-time observations. DeepMind says this yields more localized, highly accurate predictions, delivered faster.
The announcement was published as the opening post of a thread. It does not include figures on resolution, training data volume, availability terms or a numerical comparison with previous WeatherNext versions.
The rest of the details are in Google DeepMind's post on X.
