On September 3, Google introduced WeatherNext 3, a new AI weather forecasting model for users of its services, developers and researchers. According to the company, the model uses up-to-date satellite observations and generates a new forecast every hour. The update addresses both how frequently forecasts are generated and their level of detail.
Google reports a spatial resolution of 5 kilometers for near-surface temperature and humidity. This specification applies to individual variables: that level of detail cannot automatically be attributed to all WeatherNext 3 outputs. The company describes different spatial resolutions for different weather variables.
According to Google’s announcement, rollout of the model in Search, Gemini and Maps begins on the day of the announcement. The company also offers access to forecast data through queries in BigQuery and Earth Engine. This gives developers a separate way to use forecasts in their own projects.
Practical context: For an applied project, what matters here is the level of detail available for the specific variable required. For example, when selecting temperature data, the relevant forecast layer should be assessed: a model’s overall resolution figure alone does not establish whether all its data are suitable for a particular area.
Sources
Event date: 2026-09-03. Original source date: 2026-09-03.
