Google WeatherNext 3 AI Model Deployed in Search and Gemini Services, Boosting Rainfall Forecast Accuracy by 50%

Google today unveiled WeatherNext 3, positioning it as the company's "most advanced and accurate global weather AI model." The new model is immediately integrated into Search and Gemini applications, providing consumers with more detailed weather information. Previously, multiple AI weather models, including WeatherNext 2 launched in November 2025, relied on data from Numerical Weather Prediction (NWP) models for training. While NWP models are valuable, they are complex physical simulations driven by supercomputers, with data subject to a six-hour delay, often leading to inaccuracies in rapidly changing variables like rainfall or surface temperature.

WeatherNext 3 adopts a "global Earth synchronous satellite real-time data fusion" approach, allowing the model to "learn directly from real-time observations," creating a "continuously updating global atmospheric state view," and thus outputting "more timely and locally relevant weather event predictions." Developed by Google DeepMind and Google Research, this model can generate hourly forecasts at multiple spatial resolutions, covering 5-kilometer resolution for temperature and humidity, 10 kilometers for surface variables, and 25 kilometers for wind speed information.

The practical value of forecasts often hinges on their detail and the precision of their timing and spatial resolution. Compared to WeatherNext 2's specifications of a 25-kilometer grid and six-hour intervals, the new model's global weather depiction offers "about five times higher resolution." Google notes that during the sudden formation of storms, fronts, or precipitation systems, higher update frequencies and resolution can provide more detailed information in advance, aiding in the response to extreme weather. WeatherNext 3 also achieves a significant breakthrough in precipitation forecast accuracy, with its training process incorporating two sets of high-quality precipitation data, including NASA's satellite-integrated retrieval product IMERG and Google's own satellite radar-based global precipitation reanalysis data.

Google states that, based on benchmark evaluations, in medium-range global forecasts, improvements in the Continuous Ranked Probability Score (CRPS) compared to various benchmarks can reach up to 60% (compared to IMERG), 30% (compared to MRMS), and 10% (compared to rain gauges, at shorter lead times). The new model is also specifically designed for renewable energy production, capable of forecasting wind speeds at 100 meters above ground level (approximately the height of wind turbine blades) to accurately estimate wind power output, while providing high-resolution cloud cover and solar radiation data to help solar farms assess the actual sunlight received.

High-Precision Forecasting to Benefit Latin America, Africa, and the Asia-Pacific Region

Google emphasizes that WeatherNext 3 is particularly important for Latin America, Africa, and the Asia-Pacific region. These areas have long lacked high-resolution forecasting services due to the extremely high supercomputing costs required for traditional regional models. The new model brings high-quality, localized predictions to billions of people and small and medium-sized enterprises in these regions, helping to improve disaster preparedness and business decision-making. At the consumer level, WeatherNext 3 has been gradually rolled out in Google Search, Maps, and Gemini applications, with Google expecting users to see up to a 50% improvement in precipitation forecast accuracy when querying weather more than a day in advance, especially in regions where historical data shows lower forecast reliability.

The improvements are particularly significant.

WeatherNext 3 Key Specifications