WeatherNext 3 brings hourly forecasts and finer local detail to Google's weather tools
Google's new weather model uses live satellite observations for hourly forecasts, with different spatial resolutions for surface and atmospheric variables.
AI Neural Narration
48kHz StudioFish Audio Neural Engine · Natural editorial narration
Key Takeaways
- check_circleWeatherNext 3 is a forecasting model rather than a conversational LLM.
- check_circleResolution depends on the weather variable being forecast.
- check_circleEvaluate predictions against observations in your operating area.
What changed in the forecasting model
Google announced WeatherNext 3 on 3 September 2026. It uses live satellite observations and generates forecasts hourly. Temperature and moisture have a 5 km resolution; other surface variables use 10 km and atmospheric variables use 25 km.
The rollout covers Search, Gemini, Maps and Cloud services. Developers can access forecast data through BigQuery, Earth Engine and Cloud Storage. This is a weather forecasting release, so conventional LLM context and token specifications do not describe it.
Evaluate the forecast for the decision you make
Our assessment: a business should evaluate weather data against the decision it needs to make. For outdoor work, that could mean whether to move a job to another day. For deliveries, it could mean when to adjust a route. Define the useful lead time and the cost of a wrong decision before choosing an alert threshold.
Compare predictions with observed conditions over a representative period in your operating area. Separate an accurate temperature estimate from a useful prediction of rain at a particular site. Keep the forecast issue time in your records so you can judge what was knowable when the decision was made, rather than comparing against a later update.
Provide an explanation of uncertainty alongside an automated recommendation. A detailed grid is not a guarantee that conditions at an individual property will match the prediction. Keep the person responsible for the operation able to override the recommendation and record why. That makes the data useful without confusing forecast precision with certainty.
Frequently Asked Questions
Does every WeatherNext 3 variable use a 5 km grid?
No. Google specifies 5 km for temperature and moisture, 10 km for other surface variables and 25 km for atmospheric variables.
How should a business judge a forecast integration?
Assess whether it improves a defined operational decision, using predictions saved at the time and later observed conditions.