Models that learn the underlying distribution of the data and can generate new samples from this distribution.