An estimator that achieves lower variance than the Cramér-Rao bound at a specific parameter value.

An estimator that is always better than the MLE.

An estimator that converges to the true value at a rate faster than $\sqrt{n}$.

An estimator that has zero bias.