💡Easy explanation

A giant model with multiple 'experts' inside — only the relevant experts wake up for each question. More efficient because it doesn't use the whole model every time.