Google releases WeatherNext 3. The AI weather model is now more accurate thanks to live satellite data

The AI model now tracks weather directly from orbit

Google is launching WeatherNext 3, the new generation of its global AI weather forecasting model. The biggest change compared to last year's version is the shift to live satellite data. The model no longer relies solely on historical patterns; it directly analyzes the current state of the atmosphere.

Thanks to this, Google can deliver forecasts at a 5-kilometer resolution and update them hourly, which the company claims is a five times sharper picture than before. The new system is already running in Google Search, Maps, and through Gemini.

Moving away from traditional physics simulations

For meteorologists and businesses, this marks a departure from the old ways. Traditional forecasting relies on supercomputers running physical models of the atmosphere. That takes a long time and demands massive computing power. WeatherNext 3 bypasses this. It learns from massive volumes of meteorological data and finds patterns much faster.

The impact is immediate for anyone building plans around the weather (from logistics companies to solar farms and agriculture). Google is now offering much more detailed and fresher data through its API than the market is used to.

Highly localized forecasts face data quality limits

The reality check comes with the satellite data itself. While a 5-kilometer resolution sounds impressive, it only holds up where the model has access to high-quality imagery and ground sensors. In remote areas or places with poor data coverage, even the best AI will struggle to deliver accuracy.

Reliability during extreme weather anomalies is another open question. AI models excel at recognizing known patterns, but they can stumble when the weather behaves in ways that aren't common in their training data.

Independent tests and national agencies will be the judge

The key metric will be how WeatherNext 3 performs during real crises compared to the warning systems of national meteorological agencies. Google points to benchmarks (such as those from Brightband), but the true test will arrive during hurricane season or flash floods.

If the AI model proves it can spot extreme situations earlier and more accurately than traditional systems, WeatherNext 3 will become more than just a feature in Maps. It will be core risk-management infrastructure.

Lilith's verdict

The question isn’t whether you’ll see an umbrella icon in your app sooner. Whoever has faster weather data gains a market advantage before a government agency even issues a warning.