In this paper, reduced-complexity approximations of soft-output minimum mean square error (MMSE) ordering successive interference cancellation (OSIC) multiple-input multiple-output (MIMO) detector are investigated. At first, by ignoring the off-diagonal elements of the residual interference cancellation error covariance matrix and applying the matrix inverse lemma, the complexity induced by the matrix inverse is reduced drastically. Secondly, when the modulation constellation size is large, the complexity is reduced by only considering the neighboring constellation points of the estimated transmitted symbol. Finally, an effective method based on the reliability of the a posteriori symbol probability is proposed to further reduce the complexity. Extensive simulation results indicate that the proposed methods can effectively reduce the complexity at the cost of negligible performance loss.