Now your phone can tell when next storm is coming: Google’s latest weather AI model is here

Google has unveiled a significantly upgraded AI-powered weather model, announcing that it will now be integrated into widely used products such as Search, Gemini, and Pixel devices.

The company says its AI-driven forecasts have already demonstrated strong accuracy, while outperforming traditional physics-based systems in speed and efficiency. Until now, Google’s work in AI weather prediction had remained largely experimental. With this launch, the company is formally making those capabilities a core feature across its ecosystem, according to a report by The Verge.

“We’re taking it out of the lab and really putting it into the hands of users in more ways than we have before and sort of shedding off the experimental kind of designation because we have confidence that our forecasts are really quite effective and quite useful,” Peter Battaglia, senior director of research and sustainability at Google DeepMind, told reporters.

How does it work?

The new model, called WeatherNext 2, delivers forecasts up to eight times faster than its predecessor and improves accuracy for 99.9 per cent of variables, including temperature and wind.

It can produce hundreds of potential scenarios from a single starting point and generate predictions in under a minute using a Google TPU, a process that would normally take hours on a supercomputer relying on physics-based modelling.

Traditional forecasting systems require immense computational power because they simulate atmospheric physics to produce results. AI models, on the other hand, identify patterns from historical weather observations to anticipate future conditions.

How is it better than previous models?

Google says its leap in efficiency comes from the model’s architecture, which uses a Functional Generative Network (FGN). Previous AI weather models needed multiple processing cycles to build one forecast. FGN introduces controlled randomness into each input, allowing WeatherNext 2 to create numerous possible outcomes in one pass.

Thanks to these advances, WeatherNext 2 can now generate predictions up to 15 days ahead and provide hourly updates. Google expects that capability to attract interest from both consumers and enterprise customers.

“We found that energy, agriculture, transportation, logistics, and customers in many other industries are quite interested in these one-hour steps. It helps them make more precise decisions relating to things that affect their business,” said Akib Uddin, a product manager at Google Research.

Along with integrating WeatherNext 2 into Maps, Search, Gemini, and Pixel Weather, Google is also launching an early-access program for customers seeking tailored forecasting models. The data is additionally available through Google Earth Engine for geospatial work and BigQuery for large-scale analysis.

Google faces growing competition in this field, with organisations such as the European Center for Medium-Range Weather Forecasts, Nvidia, Huawei, and others developing their own AI-based forecasting systems.