Reliably predicting the weather more than a week out has been something of a Holy Grail in the meteorology industry for half a century. But a Google subsidiary now says they have built a weather model using artificial intelligence that outperforms the world's best forecasting models in terms of accuracy.

London-based Google DeepMind, which develops AI applications, published their findings in the journal Nature this week, announcing that their probabilistic weather model, called GenCast, outperformed all existing machine-learning-based forecast models in creating 15-day forecasts with great accuracy.

GenCast, the researchers say, has "greater skill and speed than the top operational medium-range weather forecast in the world," which is the ensemble forecast of the European Centre for Medium-Range Weather Forecasts, or ENS. This machine-learning model, they say, was "trained on decades of reanalysis data," and they say that GenCast "has greater skill than ENS on 97.2% of 1,320 targets we evaluated and better predicts extreme weather, tropical cyclone tracks and wind power production."

The new model uses recent advances in AI, particularly with diffusion modeling, and applies them meteorological science. GenCast was trained on 40 years of data from the European Centre, from 1978 to 2018.

As Google explains on its company blog, "GenCast is a diffusion model, the type of generative AI model that underpins the recent, rapid advances in image, video and music generation. However, GenCast differs from these, in that it’s adapted to the spherical geometry of the Earth, and learns to accurately generate the complex probability distribution of future weather scenarios when given the most recent state of the weather as input."

Speaking to the New York Times, Kerry Emanuel, a professor emeritus of atmospheric science at MIT, calls the new model and findings about its accuracy "a big deal" and "an important step forward" for atmospheric science.

DeepMind previously released a weather-forecast model in 2023 that produced accurate 10-day forecasts, but they say that GenCast was both more accurate and faster and able to take the forecasts out to 15 days with accuracy, something that Emanuel and other researchers have said would be of huge socioeconomic benefit for the planet.

"I’m a little bit reluctant to say it, but it’s like we’ve made decades worth of improvements in one year,” said the lead scientist on the GenCast project, Remi Lam, speaking to the Times. "We’re seeing really, really rapid progress."

An AI specialist at the European agency that puts out the ENS, Matthew Chantry, also tells the Times that they are already implementing some of the features of GenCast in their forecasts, saying, "That’s how highly we think of it."

GenCast isn't set to upend the weather forecasting industry overnight, but give it time. As Dr. Emanuel points out to the New York Times and the DeepMind team confirms, GenCast still isn't able to predict hurricane intensity, due to the complexities with training it on how hurricane wind speeds work. But the team believes they will get there.

"We are eager to engage with the wider weather community, including academic researchers, meteorologists, data scientists, renewable energy companies, and organizations focused on food security and disaster response," the team said in the blog post. "Such partnerships offer deep insights and constructive feedback, as well as invaluable opportunities for commercial and non-commercial impact, all of which are critical to our mission to apply our models to benefit humanity."

Related: Meta Actually Paying Celebrities Like Judi Dench and Awkwafina to Use Their Voices For Its AI Tool