Google has switched the weather forecasts in Search, the Gemini app and Google Maps over to WeatherNext 3, its latest DeepMind forecasting model. The change went live today and applies globally. You do not have to install or enable anything, and there is no subscription attached.
The headline claim is about rain. Here is how Google puts it:
When planning a day or more ahead, people will see up to 50% more accurate precipitation forecasts — with the greatest improvements in regions where forecasts have historically been less reliable.
Note the “up to.” That is Google’s own hedge, and the figure comes from Google’s own evaluation of its own model. Google does point readers to independent live leaderboards run by Brightband, but the 50% number itself is the company’s. Treat it as a claim rather than a settled result.
Where the improvement is meant to land
Google is explicit about which parts of the world stand to gain most, and this is the part that matters for readers outside North America and Europe. The company says the resolution breakthrough is particularly vital for regions across Latin America, Africa and Asia-Pacific that have historically been underserved by high-resolution forecasting, because of the supercomputing costs of running traditional regional models.
That is a real gap. Accurate short-range rain forecasting in South Asia has long been limited by how much compute a national meteorological service can afford to throw at a physics simulation. A model that runs cheaper and still resolves finer detail is genuinely useful there in a way it is not in a country already saturated with radar coverage.
What changed under the hood
The resolution numbers are easy to garble, so be precise about them. Google says WeatherNext 3 visualises key surface variables such as temperature and moisture at 5-kilometre resolution, other surface variables at 10 kilometres, and atmospheric variables like wind speed at 25 kilometres. That is not a blanket “5km forecast” — different variables sit at different grids.
Overall, Google describes the picture as roughly five times sharper than WeatherNext 2, which worked on a 25-kilometre grid in six-hour increments. In its own comparison figure, Google puts WeatherNext 2 at 25-kilometre resolution against WeatherNext 3 at 11 kilometres.
On precipitation specifically, Google trained the model on NASA’s IMERG satellite dataset and its own satellite-radar-based precipitation reanalysis. Its stated evaluation results are a Continuous Ranked Probability Score improvement of up to 60% against IMERG, 30% for MRMS and 10% against rain gauge measurements at early lead times. All three are Google’s numbers.
Google’s own caveat is the useful one
At the foot of its announcement, Google adds a disclaimer telling readers to refer to their local meteorological agency or national weather service for official forecasts, severe weather warnings and public safety advisories.
That is worth carrying forward rather than burying. The forecast in your phone’s Search results is now better at telling you whether to carry an umbrella on Saturday. It is not a cyclone warning, and Google is not positioning it as one. If a storm is bearing down, the IMD or your national service is still the source that counts.
WeatherNext 3 is also being made available to developers and researchers through BigQuery, Earth Engine and Google Cloud Storage, with hourly updates. For most people, though, the only visible change will be that the weather card in Search and Maps quietly got a new engine behind it.
