Abstract: During the severe rainstorm event in North China from 23 July to 29 July in 2025, forecast performance of CMA-EPS, EC-EPS, CMA-GFS, EC-HR, Fengqing and AIFS are evaluated using synoptic verification, threat score (TS), and MODE (method for object-based diagnostic evaluation). Results indicate that EC-EPS successfully forecasts the distribution of two rain belts with cumulative precipitation exceeding 100 mm with a lead time of 6-13 d. CMA-EPS is also able to provide probability of precipitation exceeding 100 mm 4-11 d in advance, but its forecast stability decreases as the lead time shortens. Based on MODE spatial verification, the forecasted areas of cumulative precipitation exceeding 100 mm for CMA-GFS, EC-HR, and AIFS are all smaller than observation, while Fengqing forecast area is larger. Although the centroid position of EC-HR forecasts remains relatively stable, its deviation magnitude exceeds that of the other three models. CMA-GFS model exhibit the closest agreement with the observed precipitation days in both the edge and core regions of the western Pacific subtropical high, while EC-HR model overestimates the number of precipitation days in the edge zone. Fengqing and AIFS models exhibit significant false alarms in predicting convective precipitation days for both the edge and core regions. At medium-range lead times, all four models forecast a northward displacement in the spatial distribution of heavy precipitation days compared to observations, and the forecasts shift southward approaching reality as the forecast time decreases. Additionally, forecasts from Fengqing and AIFS demonstrate advantages in forecast stability. Forecasts of CMA-GFS, EC-HR, Fengqing and AIFS models all substantially underestimate the intensity of heavy precipitation centers compared to observations, and this bias persists even as the forecast lead time decreases. Furthermore, EC-EPS ensemble members exhibit greater dispersion than CMA-EPS, with some members in long-range forecasts predicting precipitation intensities closer to observations. As the forecast lead time decreases, however, the ensemble dispersion decreases, and the predicted precipitation center intensities became even weaker than the actual conditions. Regarding the temporal evolution of heavy precipitation in the regions from Baoding of Hebei to southwestern Beijing and from northeastern Beijing to Xinglong of Hebei, EC-HR model outperforms the other three models in forecasting the intensification phase and nocturnal rainfall characteristics. Notably, Fengqing and AIFS models show a significant decline in their ability to predict precipitation intensity and nocturnal rainfall features as the forecast lead time increases. The forecasting capability for the westward extension and northward shift of the subtropical high is superior to that for its eastward retreat and southward movement. The weakened positional forecasting skill during the eastward retreat of the subtropical high may partially explain the precipitation forecast biases.