Google Launches WeatherNext 3, an AI Weather Model Using Live Satellite Data

MOUNTAIN VIEW, Calif. —

Google announced on Tuesday that it has launched WeatherNext 3, a new artificial intelligence model built to forecast weather with more detail and more frequent updates than its previous version, according to the company.

Google DeepMind and Google Research built the model, which the company said pulls in live satellite data rather than relying only on the traditional physics simulations that have long powered weather forecasting. Those older simulations, known as numerical weather prediction models, run on supercomputers but carry a six-hour lag in their data, according to Google. That lag can throw off forecasts for things that change fast, like rainfall or surface temperature, the company said.

Satellite Data Replaces the Old Six-Hour Lag

WeatherNext 3 instead reads a live mosaic of satellite images, letting it generate a fresh forecast every hour grounded in the most current observations available, Google said. The company said the model can map temperature and moisture at a resolution of five kilometers, other surface conditions at 10 kilometers, and wind speed at 25 kilometers. Google said that combination makes the new model's global forecasts roughly five times sharper than WeatherNext 2, which produced forecasts on a 25-kilometer grid updated only every six hours.

Google said the model is built on what it calls a Functional Generative Network mesh transformer, a system that produces detailed grid forecasts, tracks storm paths for cyclones, and predicts conditions at specific weather stations. The company said the added resolution should help with forecasts for areas where conditions can shift sharply over short distances, such as coastlines, valleys, and mountain ranges.

Google said WeatherNext 3 ranked as the most advanced and accurate global weather model available, citing independent live evaluations conducted by Brightband. The company did not provide independent verification of that ranking beyond citing Brightband's assessment.

Rolling Out Across Search, Maps and Cloud

Google said the model's forecasts are now integrated into Search, its Gemini AI assistant, Google Maps, Google Maps Platform, and Google Cloud, making the hourly forecasts available both to everyday users checking the weather and to businesses building their own tools on top of the data.

The company said better short-term forecasting could help with decisions in agriculture, renewable energy production, and supply chains, all of which are affected by fast-moving weather events like storms, heat waves, and droughts. Google did not disclose how many users have access to the new model or when the rollout will be complete across all of its products.