Google DeepMind and Google Research have introduced WeatherNext 3, a new global AI weather forecasting model that the company says is its most accurate yet. The practical change is not just a new research benchmark: Google is immediately wiring the model into Search, the Gemini app, Google Maps, Google Maps Platform, Google Earth Engine and Cloud-accessible data products.
For businesses that make operational decisions around rain, wind, heat and cloud cover, the announcement points to a more frequently refreshed and more granular source of forecast data without requiring them to run a weather model themselves.
What is different
WeatherNext 3 ingests real-time geostationary satellite observations alongside historical weather analysis, rather than relying only on numerical weather prediction (NWP) model outputs. Google says that enables a new global forecast every hour, compared with WeatherNext 2’s six-hour increments.
The model also increases spatial detail. It produces selected surface variables, including temperature and moisture, at 5-kilometer resolution; other surface variables at 10 kilometers; and atmospheric variables such as wind speed at 25 kilometers. Google characterizes that as roughly five times sharper than WeatherNext 2’s 25-kilometer grid.
That matters in places where terrain, coastlines and fast-moving weather can make a broad regional forecast operationally inadequate. The company says it also trains directly on sparse weather-station observations to better represent regional conditions at its global 5-kilometer grid.
A data product for energy and operations
The clearest commercial focus is clean energy. WeatherNext 3 includes forecasts for 100-meter wind speed—near turbine height—as well as high-resolution cloud cover and solar radiation. Those inputs can help wind and solar operators estimate generation, while grid operators can use expected output to balance supply and demand.
The model also targets precipitation, one of the harder variables for global forecasting systems. Google says it trained the system on NASA’s IMERG satellite precipitation data and a Google global precipitation reanalysis based on satellite radar. In its reported medium-range evaluations, the company says WeatherNext 3 improved Continuous Ranked Probability Score by as much as 60% against IMERG, 30% against MRMS and 10% against rain-gauge measurements at early lead times.
Those figures should be read as model-evaluation results, not a guarantee for a specific location or business workflow. Google itself directs users to local meteorological agencies and national weather services for official forecasts, severe-weather warnings and public-safety advisories.
Distribution is the strategic advantage
Google is making the model’s global predictions available for queries in BigQuery and Earth Engine, and for bulk download through Google Cloud Storage. That gives developers and researchers an option to incorporate weather intelligence into planning, logistics, agriculture, insurance analytics or energy workflows without first operating specialized forecasting infrastructure.
At the consumer layer, Google says WeatherNext 3 will begin powering weather experiences globally across Search, Gemini, Maps and the Google Maps Platform Weather API. It claims users planning a day or more ahead will receive up to 50% more accurate precipitation forecasts, with the largest gains in areas where forecasts have historically been less reliable.
What to watch next
The important test is how forecast quality holds up across different climates, sparse-observation regions and high-impact events—not merely on aggregate leaderboards. Operators should also examine latency, geographic coverage, licensing and how easily the data can be validated against their own local sensors and historical decisions.
Still, the release shows Google treating weather AI as both product infrastructure and a cloud data layer. The next competitive question is whether the improved resolution and hourly refreshes translate into measurable gains for businesses whose margins depend on getting the weather call right.
