Google’s latest AI weather model gives you no excuse to forget your umbrella
TechCrunch Tim Fernholz ● Covered by 4 sources
Google has a new weather AI that predicts more often and in finer detail. It’s already beating other models, and it’s headed into Search, Maps, and Gemini.
Based on reporting by TechCrunch, Tim Fernholz — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google DeepMind and Google Research have released WeatherNext 3, a new AI model for weather forecasting that the company says can see the atmosphere more clearly and predict its behavior more often. Google also plans to feed its output into Search, Google Maps and Gemini, while making it available on its cloud platforms for users and researchers.
The pitch here is not just speed. Google says WeatherNext 3 has already come out on top in tests on Operational WeatherBench, a benchmark built by the startup Brightband to compare AI forecasts. It beat other deep-learning systems from Google, Microsoft, Nvidia and ECMWF, and also outperformed traditional forecasts from the U.S. National Weather Service and ECMWF on metrics such as temperature, windspeed and humidity.
That matters because weather forecasting has been moving from giant supercomputers toward machine learning for a while now. The old systems are still very good, but they are expensive and slow. Deep learning models can work faster, and after ECMWF released more than half a century of weather data in 2018, researchers started training systems that could get close to the established tools without the same compute bill.
WeatherNext 3 tries to fix some of the weak spots that have dogged AI forecasting models. Google says it can predict key variables at 5km resolution, improve rain evaluation by 60% over WeatherNext 2, and generate hourly forecasts instead of the usual six-hour cadence. It is also bigger than its predecessor, with 2.4 times more parameters, and it targets specific weather stations rather than only broad grid averages.
The company says the model is the first AI system to directly incorporate raw observations for a high-resolution global forecast. That claim is already being nudged from the side by WindBorne, which says its WeatherMesh 6 has been doing something similar since late 2025. But even with that dispute, the direction is clear: weather AI is moving closer to the specific place and specific hour people actually care about.
My take — AI-written commentary, not fact-checked reporting
Google is doing the sensible thing here: turn a good model into a useful product instead of another demo nobody checks. The real competition in weather AI isn’t who writes the nicest paper; it’s who can make forecasts accurate enough, fast enough, and boring enough that people trust them before leaving the house.
Read more about this at: TechCrunch
