Every proof in the deep inference system MAV can be normalized to a cut-free proof using a semantic model construction.