WeatherNext 3 launches sharper, hourly forecasts
Google DeepMind and Google Research released WeatherNext 3, a global probabilistic weather model that ingests live geostationary satellite data to refresh forecasts hourly at up to 5 km resolution. It now powers Search, Gemini, Maps, and Maps Platform, while developers can request forecast datasets through BigQuery, Earth Engine, and Google Cloud Storage.
The real story is not a better weather app; it is Google turning a specialized model into an hourly, ensemble data layer for operational software. That makes WeatherNext 3 unusually relevant to developers, although real-time access remains allowlisted and experimental.
- –Live satellite ingestion enables hourly updates for fast-changing storms, compared with the six-hour cadence of many traditional global models.
- –Multi-resolution outputs reach 5 km for station-targeted temperature and humidity, 10 km for other surface variables, and 25 km for atmospheric variables.
- –Training against station, radar, and satellite observations should improve localized precipitation forecasts that earlier global AI models often blurred.
- –Native wind, cloud-cover, and solar-radiation predictions make the model directly useful for renewable-energy forecasting, logistics, agriculture, and grid planning.
- –BigQuery, Earth Engine, and Google Cloud Storage access gives developers multiple integration paths, but production systems still need independent validation and official meteorological alerts. [Google Developers](https://developers.google.com/weathernext)
DISCOVERED
1h ago
2026-09-04
PUBLISHED
7h ago
2026-09-04
RELEVANCE
AUTHOR
Sundar Pichai