This project brings together research teams to test and compare their AI climate models in a consistent way. In the first phase of the project, each model will be asked to simulate global weather patterns over several decades using the same starting conditions and boundaries like ocean temperatures and sea ice coverage. The results will be used to see how well the models can reproduce important climate features such as average temperatures, long-term trends, and natural climate swings like El Niño.

By sharing their results in a common format and comparing them using agreed-upon measurements, we hope to learn more about the strengths and weaknesses of current AI models. This will help improve future models and build trust in AI as a tool for studying Earth’s climate. The first round of results is expected in late 2025.