Google is shipping a new AI weather model that it says can predict rain and snow with up to 50 percent more accuracy at least a day in advance. The model, called WeatherNext 3, is now baked into Search, Maps, Gemini, and other Google products, putting sharper precipitation forecasts in front of a massive user base without much fanfare.
The biggest shift here is the training data. Most AI weather models learn from historical records, but WeatherNext 3 ingests real-time satellite observations and produces a forecast every hour. Google’s previous model, WeatherNext 2, updated every six hours on a 25-kilometer grid. The new one resolves temperature and moisture at up to 5 kilometers — five times sharper, according to the company. Samier Merchant, a research engineer at Google Research, framed it plainly: “One of the main developments is for [WeatherNext 3] to go beyond what data most global AI models train on.”
That speed matters most for fast-moving systems — thunderstorms, snow squalls, the kind of weather that catches people off guard. The satellite approach also helps in regions with sparse ground-based sensors, which Google says means the biggest gains will land outside the US and Europe. It’s a rare case where an AI upgrade disproportionately benefits places with less infrastructure, not more.
There’s a business angle too. WeatherNext 3 includes forecasts for renewable energy generation, including wind speed at 100 meters — roughly turbine height. Ferran Alet, a research scientist at Google DeepMind, tied it directly to the company’s ballooning data center power demands: making renewables “a very appealing opportunity” matters as energy needs climb. Google has also worked with the US National Hurricane Center and agencies in Asia to test the models.
None of this means physics-based forecasting is dead. WeatherNext 3 still trains on data from traditional supercomputer models, and agencies typically blend multiple predictions before issuing warnings. The AI is getting better at the parts weather models have always struggled with — precipitation and speed. Whether that translates into fewer soaked commuters remains to be seen, but the resolution jump alone is a meaningful signal.
