Google DeepMind’s WeatherNext 3 brings hourly, higher-resolution forecasts

Google DeepMind has introduced WeatherNext 3, a global AI weather model that uses real-time satellite observations alongside historical analysis. The Google DeepMind Blog says the system produces forecasts every hour and is now integrated into Search, Gemini, Maps, Google Maps Platform, and Google Cloud.

More detail, faster

WeatherNext 3 improves both the timing and geographic precision of its predictions. Temperature and moisture can be modeled at a 5-kilometer resolution, other surface variables at 10 kilometers, and atmospheric factors such as wind speed at 25 kilometers. That is roughly five times sharper overall than WeatherNext 2, which generated forecasts on a 25-kilometer grid in six-hour increments.

The model is designed to better capture rapidly changing, localized conditions by learning directly from live observations. The update also targets more precise rain and snow forecasts, while adding variables relevant to clean-energy planning. Google says independent live evaluations by Brightband identify WeatherNext 3 as its most advanced and accurate global model to date.

For people who build AI-powered applications, the key change is access: forecast information is not limited to a consumer weather interface. Its availability through Google Cloud and Google Maps Platform creates a path for developers to incorporate frequently refreshed, higher-resolution weather data into workflows and products. Potential use cases include agriculture, logistics, energy operations, and tools that help users respond to extreme or fast-moving weather.

The launch reflects a broader shift in applied AI: specialized models can turn continuously updated real-world data into services that are both faster and more actionable than conventional, lower-resolution forecasts.

Source: Google DeepMind Blog