Overview
WeatherNext 3 is a state-of-the-art global weather AI model jointly released by Google DeepMind and Google Research on September 3, 2026, officially described as the most accurate and advanced global weather forecasting system to date. The model became operational on August 31, 2026, building on DeepMind's landmark GraphCast work from 2023, which demonstrated that machine learning weather models can rival traditional medium-range forecasts. WeatherNext 3 directly learns from a mosaic of real-time geostationary satellite data, initializing at an hourly frequency, replacing the traditional 6- or 12-hour update cycle of numerical weather prediction, enabling finer and more timely weather insights. Its key surface variables achieve a resolution of 5 kilometers, with overall image clarity approximately five times that of the previous generation WeatherNext 2, and medium-range forecasting extends to 15 days (360 hours). The model is integrated into Google Search, the Gemini app, Google Maps, the Google Maps Platform Weather API, and Google Earth Engine, and is officially designated as the recommended default model for all new weather projects.
Key Features
- Real-time satellite data-driven, hourly updates: Directly learns from a mosaic of real-time geostationary satellite data, initializing at an hourly frequency, replacing the traditional 6- or 12-hour update cycle of numerical weather prediction, providing more immediate capture of weather conditions.
- High-resolution multi-scale output: Key surface variables achieve a resolution of 5 kilometers, other surface variables at 10 kilometers, and atmospheric variables such as wind speed at 25 kilometers, with overall image clarity approximately five times that of the previous generation WeatherNext 2.
- 15-day medium-range forecasting capability: Supports medium-range forecasts up to 15 days (360 hours), with hourly initializations outside the main synchronization cycle providing fine-grained forecasts within 48 hours, balancing timeliness and accuracy.
- Significantly improved precipitation prediction accuracy: Accuracy for one-day-ahead precipitation forecasts improves by up to about 50%, with particularly notable improvements in regions traditionally lacking high-resolution forecasts, such as Latin America, Africa, and Asia-Pacific.
- Specialized predictions for renewable energy: Can predict wind speeds at 100 meters altitude (approximately wind turbine hub height) and provides high-resolution cloud cover and solar radiation data, serving power prediction for wind and photovoltaic power generation.
- Excellent performance in independent evaluations: Independent evaluations by Brightband's Operational WeatherBench real-time assessment show an advantage of about 10% over the European Centre for Medium-Range Weather Forecasts' ECMWF AIFS ENS v2.
Use Cases
- Provides instant weather query results for Google Search, allowing users to obtain more accurate current weather and short-term forecast information.
- Integrated into the Gemini app, supporting natural language interactive weather Q&A and travel advice.
- Displays high-resolution precipitation and cloud cover layers in Google Maps, helping users plan travel routes.
- Provides enterprise-grade weather data services to developers through the Google Maps Platform Weather API.
- Supplies climate and weather analysis data to Google Earth Engine, supporting environmental monitoring and scientific research applications.
Pros
- Hourly update frequency is much higher than traditional models, reflecting weather system changes more quickly.
- 5-kilometer surface resolution provides finer local weather details.
- Significantly improved precipitation forecast accuracy, especially benefiting data-scarce regions.
- Practical renewable energy prediction capability for wind and solar power.
- Deeply integrated into the Google ecosystem, making it easy to reach a wide range of users.
- Independent evaluations show an advantage of about 10% over ECMWF AIFS ENS v2.
Pricing
The official pricing information has not been disclosed. WeatherNext 3, as the foundational model for Google's weather capabilities, is integrated into free products such as Search, Gemini, and Maps; for enterprise use of the Google Maps Platform Weather API and Google Earth Engine, specific costs are subject to Google Cloud's official pricing.
Summary
WeatherNext 3 is a global weather AI model released by Google DeepMind and Google Research in September 2026, characterized by real-time satellite data driving, hourly updates, and 5-kilometer surface resolution, providing 15-day medium-range forecasts. Its precipitation prediction accuracy improves by up to about 50%, and it supports renewable energy predictions. Independent evaluations show an advantage of about 10% over ECMWF AIFS ENS v2, and it is integrated into platforms such as Google Search, Gemini, and Maps.
Version History
- WeatherNext 3 launched and operational (2026-09-03): Google DeepMind and Google Research launched WeatherNext 3, their most advanced global weather AI model (operational since 8/31): it learns directly from real-time geostationary satellite data with hourly initialization, offers 5 km surface resolution (~5x sharper than WeatherNext 2) and 15-day forecasts. Precipitation skill improved significantly, with new clean-energy outputs such as 100 m wind speed, cloud cover and solar irradiance. Independent Brightband evaluation shows roughly a 10% advantage over ECMWF AIFS ENS v2. It now powers Google Search, Gemini, Maps, the Maps Platform Weather API and Earth Engine.