Not a single AI model is better, but what about the consensus of the 36 AI models? Ensembling different models is a common technique to improve machine learning

Not a single AI model is better, but what about the consensus of the 36 AI models? Ensembling different models is a common technique to improve machine learning models, did they test that?