Scientists use computer models to simulate Earth’s climate and predict how it might change in the future. These models are essential tools for helping governments and communities prepare for extreme weather and long-term climate shifts. Traditionally, these models rely on complex physics and decades of development. More recently, scientists have started building new types of models using artificial intelligence (AI), which learn directly from past climate data.

In this study, we evaluated how well two of these new AI-based climate models reproduce key patterns of atmospheric behavior—such as wind patterns and large-scale climate fluctuations—compared to the best traditional models. We found that in many cases, the AI models performed just as well, or even better, at capturing important features like tropical waves and shifting wind patterns in the Southern Hemisphere. However, both types of models struggled to accurately represent changes in the upper atmosphere, especially a pattern called the Quasi-Biennial Oscillation, which affects weather and climate worldwide.

Our findings show that AI-based models can already compete with the best traditional models in many areas, offering a promising path for building faster and more flexible tools to understand Earth’s climate.