Google DeepMind launches WeatherNext 3, an AI weather model using real-time satellite data for hourly, high-resolution forecasts

Google DeepMind and Google Research released WeatherNext 3, a new AI weather forecasting model that uses real-time geostationary satellite data rather than relying solely on lagged physics simulations. It produces hourly forecasts at up to 5km resolution, a sharp improvement on WeatherNext 2's 25km resolution and 6-hour update cycle, and adds renewable-energy variables such as 100m wind speed, cloud cover and solar radiation. It is already live in Google Search, Gemini, Maps, Google Maps Platform and Google Cloud, not just a research preview.

This is a genuine product upgrade with immediate distribution across Google's consumer and developer surfaces, so the practical effect is real: better short-term forecasts for hundreds of millions of users and new inputs for renewable energy planning and precipitation-sensitive industries (agriculture, logistics, insurance). Developers and enterprises get a concrete build-or-buy decision now that the model sits in Google Cloud and Maps Platform rather than staying in a lab. The claimed benefit to historically underserved regions (Latin America, Africa, Asia-Pacific) is plausible given satellite coverage but unverified by any independent regional accuracy breakdown in the source material.

The "most advanced and accurate" framing is standard Google superlative packaging and should be discounted; the only independent validation cited is a single evaluator, Brightband, with no methodology, sample size or head-to-head competitor comparison provided in the extraction. This is an incremental v2-to-v3 iteration on an existing product line, not a new capability category, despite the launch tone suggesting otherwise.

1. 01Published Brightband evaluation data or third-party benchmarks comparing WeatherNext 3 against ECMWF, NOAA and other leading forecast models over the next 1-3 months
2. 02Actual developer adoption on Google Cloud and Maps Platform (API usage, pricing, named enterprise customers) rather than announcement-stage availability
3. 03Independent accuracy checks for underserved regions (Latin America, Africa, Asia-Pacific) specifically, since satellite-based input quality can vary by ground station density and cloud cover in those areas