The belief propagation (BP) algorithm is inherently parallel with low latency. Nevertheless, especially in high SNR, the error correction capability of BP is uncompetitive. To lower the block error rate (BLER), this study proposes a multi-bit flipping algorithm based on dynamic flipping matrix. The polarized channel’s transmission error probability is analyzed and the most unreliable channels are chosen as the static flipping positions. When conventional BP decoding encounters failure, We determine the dynamic positions by selecting the log-likelihood ratio (LLR) with the smallest absolute value. Finally we combine the static positions and dynamic positions to construct a dynamic flipping matrix. The numerical results show that compared with the bit-flipping BP (BFBP)-CS- \({\omega ^3}\) , with the assistance of the dynamic flipping matrix, the proposed bit-flipping BP has a performance gain of 0.13 dB at BLER \( = \) \( {10^{-3}}\) . Compared with the BFBP-CS- \({\omega ^2}\) , the complexity is reduced by 22.27% at Eb/N0 \( = \) 3 dB.