Abstract: Summer rainstorms on the northern slope of the Kunlun Mountains is highly sudden. This makes forecasting and early warning very difficult. The performance of model precipitation forecast products in predicting rainstorm in this region remains unclear. Based on precipitation observation data and multi-source merged precipitation products, this study combines traditional point-to-point verification with the MODE method to evaluate and compare the forecast performance of ECMWF, CMA-GFS, and CMA-MESO models for three rainstorm events on the northern slope of the Kunlun Mountains during summer 2021-2022, assessing both precipitation intensity and spatial distribution.The results are as follows. (1) All three models tend to overestimate light and moderate rainfall but underestimate heavy rain and rainstorms. The ECMWF model performs better in capturing moderate and heavy rain, while CMA-MESO shows superior skill in rainstorm forecasting, outperforming CMA-GFS. (2) During intense and weak rainfall phases of rainstorm events, all three models show higher false alarm and miss rates; this underestimation of torrential rain intensity primarily arises from the systematic underestimation of precipitation amounts coupled with overestimation of precipitation frequency. (3) ECMWF outperforms CMA models in matching 24-hour and 3-hour accumulated precipitation objects with observations. ECMWF’s 24-hour precipitation objects show a southward and eastward displacement, while CMA models exhibit a southward and westward bias. All models effectively capture the morphology of precipitation objects but consistently underestimate their coverage area.