> ## Content Index
> Fetch the complete content index at: https://globalfeed.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# Google DeepMind launches WeatherNext 3 with hourly, 5-kilometre forecasts
- URL: https://globalfeed.ai/en/google-deepmind-launches-weathernext-3-with-hourly-5-kilometre-forecasts/
- Published: 2026-09-04T09:28:38.000Z
- Updated: 2026-09-04T09:28:38.000Z
- Description: The new model learns from live geostationary satellite data, refreshes every hour and is already running behind Google Search, Maps and Gemini.
- Author: GlobalFeed Editor
- Tags: Google DeepMind, WeatherNext, weather forecasting, x-GoogleDeepMind, dil-en, denetim-gerekli, video, x-gitti

Google DeepMind and Google Research have released **WeatherNext 3**, a forecasting model that learns directly from live geostationary satellite imagery rather than from the output of physics-based simulations. It produces a new forecast every hour. Google says independent live evaluations by Brightband rank it as the most accurate global weather model to date.

## Five times sharper

WeatherNext 3 resolves surface variables such as temperature and moisture at 5 kilometres, other surface variables at 10 kilometres and atmospheric variables such as wind speed at 25 kilometres. The previous version, WeatherNext 2, ran on a 25-kilometre grid in 6-hour increments, so the overall picture is roughly five times sharper.

The change comes from the training data. Most AI weather models, WeatherNext 2 included, are trained on numerical weather prediction (NWP) output, which is generated by supercomputer physics simulations and carries a six-hour lag. Google says that lag introduces bias in fast-moving variables such as rain and surface temperature. The new model also trains directly on sparse weather station observations, which the company says helps it capture local variation near coastlines, valleys and mountain ranges.

The system runs on a single Functional Generative Network (FGN) mesh transformer. Alongside dense gridded fields, it outputs discrete cyclone tracks and station-level point predictions.

## Energy variables

The model forecasts wind speed at 100 metres, roughly turbine height, for wind-energy output estimates. It also produces high-resolution cloud cover and solar radiation levels for solar farms.

Google notes that high-resolution forecasting has been limited across Latin America, Africa and Asia-Pacific because regional models carry heavy supercomputing costs, and says the new model extends coverage to those areas. WeatherNext 3 is now integrated across Google Search, Gemini, Maps, Google Maps Platform and Cloud.

Full details are in the [Google DeepMind announcement](https://deepmind.google/blog/introducing-weathernext-3-our-most-advanced-and-accurate-global-weather-ai-model/?ref=globalfeed.ai).