Analysis
Google DeepMind rolled out WeatherNext 3 this week, calling it the company's most advanced and accurate global AI weather model yet, according to Google's own announcement and 9to5Google. The model generates forecasts every hour on a 5-kilometer grid -- up from WeatherNext 2's 25-kilometer resolution -- and DeepMind says it delivers up to 50% more accurate precipitation forecasts a day or more in advance.
The technical shift underneath the accuracy claim is training methodology: WeatherNext 3 trains directly on raw satellite observations rather than the output of traditional numerical weather models, removing a six-hour data-refresh lag that has constrained forecasting accuracy for decades regardless of how much compute gets thrown at the problem. The model now also forecasts wind speeds at 100 meters -- roughly turbine height -- along with cloud cover and solar radiation, data DeepMind says can help grid operators estimate renewable power output more precisely.
Google's second weather model in a year, now with an energy angle
WeatherNext 3 immediately powers weather results across Google Search, Gemini, Google Maps, and the Maps Platform Weather API -- distribution that dwarfs what any standalone weather-AI startup can match. The renewable-energy forecasting angle is the more commercially interesting expansion: accurate wind and solar output prediction is a genuine pain point for grid operators managing intermittent renewable capacity, putting Google in loose competition with specialized energy-forecasting vendors, not just consumer weather apps.
What DeepMind hasn't published alongside the launch is independent, third-party validation of the 50% accuracy claim against a standardized benchmark like the ECMWF's own verification metrics -- the number comes from DeepMind's internal testing, the same caveat that applies to most AI weather-model claims industry-wide, including Nvidia's FourCastNet and Huawei's Pangu-Weather. Whether WeatherNext 3's advantage holds up once meteorologists outside Google run their own comparisons is the open question before it becomes the new industry baseline rather than a marketing number.
