Google DeepMind and Google Research Launch WeatherNext 3 AI for Hourly Global Forecasts, Powering Google Search and Maps

Google DeepMind Unveils WeatherNext 3 with Live Satellite Integration

Google DeepMind and Google Research introduced WeatherNext 3 on September 3, 2026, as their most advanced global AI weather model, which learns directly from live satellite data to provide hourly forecasts. This new model replaces physics-simulation inputs with AI forecasts, generating predictions at up to 5-kilometer resolution, a five-fold increase in sharpness over WeatherNext 2. For broader context, explore our AI News.

WeatherNext 3 is already integrated into various Google services, including Google Search, the Gemini app, Google Maps, the Google Maps platform Weather API, and Google Earth Engine. This integration provides users with more precise and frequently updated weather information.

Direct Satellite Data: A New Approach to AI Weather Forecasting

Unlike previous AI weather models that were trained on outputs from physics-based simulations, WeatherNext 3 learns directly from raw observational data. This fundamental change allows the model to process live geostationary satellite mosaics as a direct input, enabling it to produce a new forecast every hour. This capability makes it the first global weather model to offer such frequent updates, initialized directly from real-time satellite observations.

The architecture behind WeatherNext 3 is a Functional Generative Network (FGN) mesh transformer, which facilitates its ability to learn complex weather patterns from vast amounts of live data.

Enhanced Accuracy and Resolution for Global Forecasts

WeatherNext 3 delivers substantial improvements in forecast accuracy, particularly for precipitation. It provides up to 50% more accurate precipitation forecasts for day-ahead planning, with notable gains in regions that have historically been underserved by traditional forecasting methods. The model's ability to generate forecasts at a 5-kilometer resolution represents a significant leap from the 25-kilometer grid and 6-hour increment cycle of earlier models, offering a much sharper and more localized view of weather conditions.

This enhanced resolution and hourly refresh rate address a critical need for timely and precise weather information, reducing the typical six-hour data lag associated with some forecasting systems.

Supporting Renewable Energy and Broader Applications

Beyond general weather forecasting, WeatherNext 3 incorporates new variables crucial for the renewable energy sector. It includes predictions for 100-meter wind speeds and solar irradiance, which are vital for optimizing output estimates for grid operators and both wind and solar farms. This addition underscores the model's utility in supporting critical infrastructure and the transition to cleaner energy sources.

The data generated by WeatherNext 3 is also made accessible to researchers and businesses. It can be accessed hourly via BigQuery, Earth Engine, and through bulk downloads from Google Cloud Storage, facilitating further analysis and application development.

Key Integrations and Official Advisories

WeatherNext 3's integration across Google's ecosystem means that its advanced forecasts are now powering everyday tools. Users of Google Search, the Gemini app, and Google Maps will benefit from the improved accuracy and timeliness of weather information. Developers utilizing the Google Maps Platform Weather API and researchers working with Google Earth Engine also gain access to this enhanced data.

Google emphasizes that national weather services remain the authoritative source for official warnings and safety advisories, highlighting the complementary role of WeatherNext 3 in providing detailed forecast data.

Conclusion

Google DeepMind and Google Research's introduction of WeatherNext 3 marks a significant advancement in AI-driven weather forecasting. By leveraging live satellite data and an hourly refresh cycle, the model offers unprecedented resolution and accuracy, particularly for precipitation and renewable energy variables. Its integration into widely used Google products and accessibility for researchers and businesses positions WeatherNext 3 as a foundational tool for future weather-dependent applications, while reinforcing the importance of official weather services for critical safety information.

Sources

• WeatherNext 3 — Google DeepMind
• [2609.03582] WeatherNext 3: Increasing resolution and performance of global weather models with raw observations
• WeatherNext 3: Our most advanced global weather AI model
• WeatherNext 3: Increasing resolution and performance of global...
• AI model achieves breakthrough in forecasting cyclones