Based on observations and BCC-CPSv3 model data from 2023-2024, this study employs a multidimensional evaluation framework of 'region-season-time' integrating metrics such as bias, root mean square error (RMSE), time correlation coefficient (TCC), and spatial anomaly correlation coefficient (ACC), and log transformation to systematically analyze the model's multi-scale prediction skills and error characteristics for temperature and precipitation in Inner Mongolia. Key findings include: Temperature predictions exhibit significant seasonal biases (a warm bias of +3.1℃ in summer over eastern farmland areas and a cold bias of -2.2℃ in spring over western deserts). Prediction skill (TCC) is markedly higher in winter (TCC=0.43) than in summer (TCC=0.28), with spatial heterogeneity (higher in east, lower in west) driven by underlying surface variations. Precipitation predictions are dominated by extreme values (daily precipitation >20 mm), showing a spring positive bias (+0.64 mm) in the west and a summer negative bias (-0.24 mm) in the east. Log transformation effectively isolates 58% of extreme-value-induced errors. Prediction skill peaks in winter with gradual decay (-0.04 per ten-day period), while other season's exhibit abrupt declines. Spatial distribution follows a 'higher in west, lower in east' pattern. Mechanistic analysis reveals opposing spatial patterns between temperature and precipitation skills, temperature depends on large-scale circulation stability (e.g., winter Mongolian High dominance), whereas precipitation is constrained by convective intermittency (summer meso-microscale systems). The study identifies targeted optimization pathways: adjusting latent heat parameters for eastern farmland to correct temperature biases; modifying convective triggering thresholds in summer and westerly moisture simulations in spring to improve precipitation predictions. These findings provide critical insights for mitigating spring droughts, summer floods, and advancing regional model refinement in Inner Mongolia.