Google’s latest AI weather model gives you no excuse to forget your umbrella

The landscape of meteorology is undergoing a seismic shift, driven by the rapid evolution of deep learning. Today, researchers at Google DeepMind and Google Research unveiled their latest breakthrough: WeatherNext 3. This advanced artificial intelligence model represents a significant leap forward in how we observe, interpret, and predict the chaotic behavior of our atmosphere.

Beyond the laboratory, this technology is set to become a household utility. Google has confirmed that WeatherNext 3 will soon power the weather information displayed across its most popular platforms, including Google Search, Google Maps, and Gemini. Furthermore, the model will be accessible to researchers and developers via Google’s cloud infrastructure.

“This is going to be the first time that some of the core variables feed and power a lot of the Google products,” noted Samier Merchant, a senior staff engineer at Google.

Setting a New Benchmark for Accuracy

In rigorous testing conducted on Operational WeatherBench—a specialized utility developed by the startup Brightband—WeatherNext 3 has emerged as the top performer. By analyzing critical metrics such as temperature, humidity, and wind speed, the model has outperformed a host of formidable rivals.

Its success isn't limited to other deep-learning models from tech giants like Microsoft and Nvidia; it has also surpassed the predictive capabilities of traditional, physics-based forecasts generated by the U.S. National Weather Service and the European Center for Medium-Range Weather Forecasting (ECMWF).

The Evolution of Forecasting: From Supercomputers to AI

For decades, weather forecasting has relied on massive, government-owned supercomputers. These machines laboriously process complex mathematical equations to simulate the physics of the atmosphere. While highly accurate, these systems are notoriously expensive to maintain and relatively slow to compute.

The paradigm shifted in 2018 when the ECMWF released over 50 years of historical weather data. This treasure trove allowed deep learning researchers to train models capable of generating predictions at a fraction of the time and cost, while maintaining—or exceeding—the accuracy of traditional methods.

“Weather is chaotic, and so small differences really start to perturb massively… Machine learning targets the problem we are really solving, which is approximate noisy physics from incomplete information and finite compute, and so it learns patterns from a lot of data,” explained Ferran Alet, a staff research scientist manager at DeepMind.

Solving the "AI Weather" Problem

Despite the promise of AI, early models faced three persistent hurdles: they were often too coarse (forecasting over 15–25 square km), struggled with precipitation accuracy, and remained tethered to pre-formatted government datasets. WeatherNext 3 addresses these limitations head-on:

• Granular Resolution: The model can now predict weather patterns down to a 5 km resolution.
• Enhanced Precipitation Tracking: Evaluations show a 60% improvement in rain forecasting compared to its predecessor, WeatherNext 2.
• Increased Frequency: The model now delivers hourly forecasts, a major upgrade from the industry-standard six-hour intervals.

These advancements are largely due to a larger architecture, featuring 2.4 times more parameters than previous iterations. Additionally, the researchers tuned the model to visualize specific phenomena, such as cyclone paths, and trained it to target specific ground-truth weather stations.

“The idea, with a lot of AI applications, is to try to run tasks as end-to-end as possible,” said Daniel Rothenberg, an atmospheric scientist at Brightband. “Adding a capability where this model is now also predicting, say, what Denver’s airport’s weather station is going to measure on an hourly basis, just connects that forecasting task closer to the core.”

The Future of Global Data Assimilation

A standout feature of WeatherNext 3 is its ability to ingest raw, real-time satellite data on an hourly basis. While Google claims this is the first AI model to directly incorporate such raw observations for high-resolution global forecasting, the field is competitive. The startup WindBorne has been utilizing its own fleet of weather balloons to feed raw data into its WeatherMesh 6 model since late 2025. Google maintains that its own solution offers superior global resolution.

Regardless of the competition, the industry is moving toward a future where AI-driven meteorology provides tangible economic benefits. From helping farmers in developing nations improve crop yields to optimizing renewable energy projects by providing more reliable wind and cloud cover data, the impact is profound.

As Ferran Alet summarized, the ultimate goal remains the same: “At the end of the day, I think Google is about providing useful information to the user, and a lot of what users are looking for has to do with the weather in some way or another.”

With WeatherNext 3, that information is about to become faster, sharper, and more reliable than ever before.