We develop new theoretical guarantees for optimal transport map estimation under minimal regularity assumptions, providing the first provable error bounds for neural-OT estimators between general distributions.