Model-based learning involves building or using an approximate model of the environment's transition probabilities, while model-free learning directly estimates values and policies without using the transition probability distribution. 

Model-based learning relies on trial-and-error to learn optimal actions, while model-free learning computes exact solutions based on a known model of the environment. 

Model-based learning assumes the environment's transition dynamics are deterministic, while model-free learning assumes the environment's transitions are probabilistic.