In reinforcement learning and related fields, a _reward model_ is a function that assigns a scalar value (a reward) to a given state, representing how desirable it is. You're at liberty to have compound states: for an example, a trajectory (often called tau) or a state action pair (typically represented by s and a).