Google has begun rolling a new forecasting model into Search, Maps and Gemini, and the early evidence gives it something more useful than a new name. In Google’s tests, WeatherNext 3 reduced one measure of short-range temperature error by up to 30% compared with WeatherNext 2. That is encouraging news for anyone making plans around the temperature forecast.
Here’s what that percentage actually measures. The model predicts a range of possible temperatures and how likely they are. Researchers score how well that picture matches the temperatures that occur. A lower score means a better probabilistic forecast. The 30% result applies to that particular short-range test, using weather stations excluded from training; it is not a promise that every forecast is 30% better.
One practical improvement is how often fresh information arrives. WeatherNext 3 takes in live satellite observations and starts a new forecast every hour. Its main runs, six hours apart, look up to 15 days ahead; the updates between them cover the next 48 hours. When conditions change quickly, newer observations give the model more to work with.
It also produces 64 possible weather scenarios. That helps describe uncertainty, which is useful when a single confident-looking number would hide the range of things that could happen. The model’s outputs include wind at turbine height and sunlight, making them relevant to people planning renewable-power generation as well as people planning a walk.
There is encouraging outside evidence, too. Brightband’s independent global rankings put WeatherNext 3 first across the seven weather variables scored in the October 7 snapshot. Its current comparison does not score this model’s rainfall, so that result cannot validate every claim about rain. Google’s research reports separate rainfall improvements against observation-based tests.
Google announced the consumer rollout on September 3. The everyday payoff is better information in tools people already check: how cold it may get, how conditions could change and how much confidence to put in the outlook. For dangerous weather, official alerts from national weather services and local authorities still come first.
