Landslides on gently inclined loess–bedrock contact surfaces are common geological hazards in the northwestern Loess Plateau region of China and pose a serious threat to the lives and property of local residents as well as sustainable regional development. Taking the Libi landslide in Shanxi Province as a case study (with dimensions of 400 m × 340 m, maximum thickness of 35.0 m, and volume of approximately 3.79 × 104 m3, where the slip zone is located within the highly weathered sandy mudstone layer of the Upper Shihezi Formation of the Permian System), this study employed a combination of physical model experiments and numerical simulations to thoroughly investigate the formation mechanism of gently inclined loess landslides. Via the use of physical model experiments, a landslide model was constructed at a 1:120 geometric similarity ratio in addition to three scenarios: rainfall only, rainfall + rapid groundwater level rise, and rainfall + slow groundwater level rise. The dynamic changes in the water content, pore water pressure, and soil pressure within the slope were systematically monitored. Numerical simulations were conducted via GEO-STUDIO 2012 software to further verify and supplement the physical model experimental results. The research findings revealed that (1) under rainfall conditions alone, the landslide primarily exhibited surface saturation and localized instability, with a maximum displacement of only 0.028 m, which did not lead to overall instability; (2) under the combined effects of rainfall and rapid groundwater level rise, a “sudden translational failure mode” developed, characterized by rapid slope saturation, abrupt stress adjustment, and sudden overall instability; and (3) under conditions of rainfall and a gradual groundwater level rise, a “progressive translational failure mode” emerged, experiencing four stages: initiation, development, acceleration, and activation, ultimately resulting in translational sliding of the entire mass. Through a comparative analysis of physical model experiments, numerical simulation results, and field monitoring data, it was verified that the Libi landslide belongs to the “progressive translational failure mode”, providing important theoretical basis for the identification, early warning, and prevention of such types of landslides.