Google DeepMind and Google Research have released WeatherNext 3, an AI weather forecasting model that will power weather information across Google Search, Maps, and Gemini. The model outperforms competitors on benchmarks and offers higher resolution and hourly forecasts, marking a significant advancement in AI-driven weather prediction.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Google DeepMind has released WeatherNext 3, claiming it is its most accurate AI weather model, and for the first time, core forecast variables are being integrated directly into Search, Maps, and Gemini. But where are the boundaries of this breakthrough?
On September 4, Google DeepMind and Google Research jointly released WeatherNext 3, announcing that the model will begin providing weather information to Search, Maps, and Gemini, while also being made available to researchers on the cloud platform. Senior Staff Engineer Samier Merchant stated that this is the first time core forecast variables have directly supported Google's major products.
This move signifies that AI weather forecasting has evolved from a research tool to infrastructure serving billions of users, with an impact far beyond the academic sphere.
WeatherNext 3 achieves 5 km resolution on key variables, improves rainfall prediction by 60% over the previous generation, and can generate hourly forecasts instead of the standard six-hour intervals. These advances are partly due to its ability to ingest real-time satellite data.
On the Operational WeatherBench benchmark, built by startup Brightband, the model reportedly surpasses leading competitors such as Microsoft, NVIDIA, and the European Centre for Medium-Range Weather Forecasts, as well as traditional forecasts from national weather agencies.
It should be noted that these performance claims come primarily from Google's official statements, and independent verification has not yet been presented in available materials.
Ferran Alet, Research Scientist Manager at DeepMind, explains that machine learning approximates chaotic, noisy physical processes that traditional systems struggle to compute efficiently by learning patterns from vast amounts of data. This explains why AI models can achieve breakthroughs in computational efficiency and accuracy.
Daniel Rothenberg of Brightband emphasizes that the model's new ability to tailor forecasts to specific weather stations brings AI predictions closer to real-world ground-truth applications, such as predicting conditions at a specific airport.
Google claims WeatherNext 3 is the first high-resolution global model to directly incorporate raw satellite observations, but competitor WindBorne has used similar raw data technology since late 2025. This claim therefore faces a precedent challenge.
This controversy reminds us that in the rapidly evolving field of AI weather forecasting, the label of 'first' often requires careful interpretation, and judgments of technological leadership may depend on definitions and scope.
This article is primarily based on reporting from The AI Insider, where performance data and product plans are largely from Google's official statements, falling under the source_claim level; independent verification has not yet been presented in available materials.
WeatherNext 3 marks the transition of AI weather forecasting from research to mainstream products, but independent verification of its performance advantages, the controversy over the 'first' claim, and the actual deployment effectiveness still require further observation.
Primary report
Primary source
Also covered by 1 sources