Saturated hydraulic conductivity (Ks) is a critical parameter for assessing water-induced loess collapsibility, erosion, and landslides. However, accurately determining Ks has long been a challenge in geological and geotechnical engineering due to the complexity and inherent spatial variability of loess-paleosol sequences. To address this issue, this study conducted shaft sampling and laboratory experiments to measure the Ks of loess with a deposition time (T) of up to 880 ka. By leveraging the well-defined deposition time scale and global relevance of loess, a predictive model incorporating Ks variability was developed with T as a variable. This paper provides a detailed discussion of the physical significance of the model's parameters, their determination methods, and verifies its applicability. Pore distribution and scanning electron microscope (SEM) images were used to reveal the three-stage evolution of Ks over time, as well as the underlying microstructural mechanisms. Additionally, this paper explores the impact of commonly used merging layer methods on Ks variability in engineering practice. The model effectively captures the long-term evolution of Ks in loess and can predict the Ks of loess-paleosol sequences, along with their expected variability, at a lower cost. This provides more reliable parameters for geological hazard assessments and hydrological engineering design.