This study applies a non-stationary bias correction method, integrating techniques of quantile delta mapping and rank resampling for distributions and dependencies, to evaluate and enhance precipitation simulations from 23 CMIP6 models over the Huang-Huai-Hai and Jiang-Huai plains using daily gridded precipitation datasets, examining model performance in reproducing both climatological mean precipitation patterns and extreme precipitation characteristics before and after bias correction. The results show that, prior to correction, the CMIP6 multi-model ensemble demonstrates systematic overestimation of mean annual precipitation across both plains, characterized by underestimation in summer and overestimation in other seasons, while regarding extreme precipitation events, the uncorrected ensemble exhibits overestimation of precipitation frequency and maximum consecutive wet periods alongside underestimation of precipitation intensity and maximum consecutive dry periods. The implementation of non-stationary bias correction method yields substantial improvements in CMIP6 precipitation simulation capabilities, with post-correction results demonstrating climatological mean precipitation biases constrained within ±10% and average TS scores for eight extreme precipitation indicators enhanced by approximately 10%.