Introducing WeatherNext 3, our most advanced and accurate global weather AI model

Details

WeatherNext 3 marks a concrete shift in how Google builds weather models. Rather than relying solely on historical weather records, the new model ingests real-time satellite observations, enabling hourly forecast updates where its predecessor, WeatherNext 2, only produced outputs every 6 hours. The resolution improvement is substantial: WeatherNext 2 operated on a 25-kilometer grid; WeatherNext 3 visualizes temperature and moisture at up to 5-kilometer resolution — a 5x sharpness increase. Google claims precipitation accuracy improves by up to 50% for forecasts at least one day ahead, driven by the model's ability to track fast-moving weather systems with live satellite data. Coverage gaps in data-sparse regions are a direct target. Outside the US and Europe — where ground-based rain gauges are sparse — satellite-derived predictions fill in what physical sensors miss. Google says WeatherNext 3 delivers its biggest accuracy gains in these underserved regions. Enterprise use cases are built in: the model includes clean energy planning variables (wind, solar) and is accessible via Google Cloud for developers building agriculture, logistics, and renewable energy applications. WeatherNext 3 is also now live in consumer-facing products: Search, Maps, and Gemini.