Google’s latest AI weather model gives you no excuse to forget your umbrella — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Google DeepMind released WeatherNext 3, an AI weather model that achieves 5 km resolution (down from 15–25 sq km in prior AI models) and produces hourly forecasts instead of every six hours.
- WeatherNext 3 topped Operational WeatherBench benchmarks, beating AI models from Microsoft, Nvidia, and the ECMWF as well as traditional forecasts from the U.S. National Weather Service.
- Rain predictions improved 60% over WeatherNext 2, driven by a model with 2.4x more parameters and training targeted at specific weather data stations for ground-truth validation.
- Google says WeatherNext 3 is the first AI model to directly incorporate raw satellite observations into a high-resolution global forecast, though startup WindBorne's WeatherMesh 6 has done so since late 2025; Google countered that its resolution is higher globally.
- WeatherNext 3 will feed weather data into Google Search, Google Maps, and Gemini — the first time its core variables power Google's consumer products, according to senior staff engineer Samier Merchant.
- DeepMind staff research scientist manager Ferran Alet framed the advance as machine learning targeting "approximate noisy physics from incomplete information," arguing AI models sidestep the cost wall of government supercomputers and could extend forecasts to poorer regions.
Why it matters: WeatherNext 3 is the first Google weather model whose core outputs actually flow into Search, Maps, and Gemini, giving Google a differentiated AI feature across consumer surfaces at a fraction of the cost of government supercomputers — a gap the source explicitly identifies as the route to bringing accurate forecasts to poorer regions that can't afford traditional infrastructure.
Ask SkimNews
.gif)
