Google's AI weather model gets a satellite upgrade, here's what changes
WeatherNext Gen 3 now pulls in raw satellite data, which means fresher forecasts that cost far less to run than traditional weather models.
Key points
- Google released WeatherNext Gen 3 in 2025, the third version of its AI-driven weather forecasting model.
- The biggest update: the model now ingests satellite weather data directly, cutting the delay between real-world conditions and a new forecast.
- AI weather models can match traditional forecast accuracy while using a fraction of the computing power, allowing them to run more often.
- The update is backed by a technical white paper published by Google.
Google has quietly become one of the biggest names in AI weather forecasting, and its latest move makes its forecasts faster and fresher.
The company released WeatherNext Gen 3, the newest version of its machine-learning weather model. A machine-learning model, at its simplest, is software that finds patterns in huge amounts of data and uses those patterns to make predictions.
What actually changed in this new version?
The headline improvement is satellite data. WeatherNext Gen 3 now pulls in readings directly from weather satellites, cutting the lag between what the atmosphere is doing right now and when a forecast gets generated.
Previous versions leaned more heavily on something called a reanalysis: a global atmospheric snapshot that takes weather readings from thousands of sources, fills in gaps over oceans and remote areas, and stitches everything into one consistent picture a forecast model can use. That process takes time. Going straight to satellite data shortens that wait.
Google has published the technical details in a white paper. Our 3 September story found that WeatherNext 3 checks live satellite images every hour and can spot a storm forming over areas that barely have a rain gauge.
Why should ordinary people care about any of this?
AI weather models do something traditional supercomputer-based forecasting struggles with: they deliver comparable accuracy at a fraction of the computing cost. Lower cost means the model can run more often and push more frequent forecast updates throughout the day.
For a nurse checking whether a thunderstorm will hit during her commute, or a market trader watching crop-weather risk, that frequency is the point.
Here is a quick look at where WeatherNext sits in the broader field:
| Model | Developer | AI-based? | Satellite data in Gen 3? |
|---|---|---|---|
| WeatherNext Gen 3 | Yes | Yes (new) | |
| GraphCast | Google DeepMind | Yes | Partial |
| Pangu-Weather | Huawei | Yes | No |
| ECMWF AIFS | ECMWF | Yes | Yes |
| GFS | NOAA | No | Yes |
As first reported by Ars Technica, the satellite integration is the sharpest technical edge in this release.
What happens next?
AI weather models are improving faster than traditional models did over a comparable period. Closing the remaining gaps around extreme events like hurricanes matters enormously: every hour of lead time saves lives.
No AI model has fully replaced the big operational systems yet. Each generation narrows the gap, though, and WeatherNext Gen 3 is a real step forward.
The honest takeaway: If your work or safety depends on weather forecasts, watch whether your app or service has started using AI-based models. Fresher, more frequent updates are coming, and they won't cost you a thing.
Common questions
Does this mean weather forecasts will be more accurate starting now?
Not overnight for consumers. The improvement is mainly in freshness and update frequency, not a dramatic jump in accuracy for everyday forecasts. Accuracy gains are incremental and vary by weather type.
Is Google selling this forecast data, or is it free?
WeatherNext is primarily a research and infrastructure product. A paid consumer weather service hasn't been announced, though the technology feeds into broader products over time.
