Abstract AI climate model simulations are in their infancy and the paper by Baxter et al. (this issue) reports the results of a new set of tests for how well AI weather and climate models represent aspects of atmospheric dynamics, including low frequency variability. They shine a light on previously untested aspects of AI models and highlight their very mixed performance, illustrating some stunning successes and some disappointing failures. This kind of work is an important prerequisite for reliable longer term climate prediction with AI models and there are implications for training future AI models.