What happened
TechCrunch reports that Google DeepMind and Google Research released WeatherNext 3, an AI weather-forecasting model that will begin supplying weather information for Google Search, Maps, and Gemini, while also being made available to users and researchers through Google’s cloud platforms.
TechCrunch reports that WeatherNext 3 can forecast key variables, including temperature, wind speed, and humidity, at a resolution as fine as 5 kilometers. The outlet also reports that its rain evaluations improved by 60% compared with WeatherNext 2 and that the model can generate hourly forecasts rather than the more common six-hour intervals. Google has not documented specific consumer availability or pricing in the source.
According to TechCrunch, WeatherNext 3 is 2.4 times larger than its predecessor and was trained to forecast conditions at specific weather stations, helping connect predictions with ground-based observations. The model also ingests weather-satellite data collected hourly. TechCrunch reports that Google describes this as the first AI model to directly use raw observations for a high-resolution global forecast, although WindBorne disputes that distinction and says its WeatherMesh 6 model has used raw observations since late 2025. The source says both systems still rely on national weather datasets, so full direct data assimilation remains unresolved.
Source details: techcrunch.com ↗
Why it matters
Higher-resolution and more frequent forecasts could make AI-based weather information more useful for local planning, renewable-energy operations, agriculture, and regions that cannot afford the infrastructure traditionally used for numerical weather prediction. The reported results remain claims from Google and benchmark evaluations described by TechCrunch, not independent confirmation.
The reported changes address several practical weaknesses of AI weather systems: coarse geographic resolution, weaker precipitation forecasts, infrequent updates, and reliance on preprocessed government datasets. More localized and frequent predictions could improve decisions that depend on wind, rain, and cloud cover, including renewable-power planning and agricultural operations.
TechCrunch says WeatherNext 3 performed better than other deep-learning models from Google, Microsoft, Nvidia, and the European Centre for Medium-Range Weather Forecasts, as well as traditional forecasts from the U.S. National Weather Service and ECMWF, on Operational WeatherBench. Those results are not independently confirmed here, and the source does not provide the underlying scores, test periods, geographic breakdowns, or details needed to assess how broadly the advantage applies.
What to watch next
Watch for the model’s actual rollout across Google products, the terms and pricing for cloud access, and independent evaluations of performance in difficult local rainfall and extreme-weather conditions.
The key practical question is access. TechCrunch reports that WeatherNext 3 will feed Google products and be available through Google Cloud, but the article does not establish when each integration will appear, which users will receive it, what APIs or services are involved, or what cloud access will cost.
Further scrutiny should focus on heavy-rain events, cyclones, regional performance, forecast calibration, and whether the model’s higher resolution translates into better decisions rather than merely more detailed outputs. The distinction between raw-observation ingestion and genuine direct data assimilation also remains an important technical limitation.