Based on the hindcast datasets from the Fengshun Model and the S2S-ECMWF Model, together with CMA-RA1.0 and NCEP reanalysis data and station observations during 2017-2021, the pentad-scale prediction skills of the Fengshun Model and the S2S-ECMWF Model for midsummer (July-August) temperature over China are comparatively verified. Three metrics of temporal correlation coefficient (TCC), anomaly correlation coefficient (ACC), and integrated prediction score (IPS) are adopted. The results are as follows. The verification results based on different reference datasets are generally consistent, demonstrating their good robustness. Overall, the Fengshun Model shows superior prediction skill to the S2S-ECMWF Model, with TCC, ACC and IPS values improved by 7.9%, 18.4%, and 1.5%, respectively. Spatially, the Fengshun Model exhibits higher TCC skill in the Huang-Huai (Yellow River-Huaihe River), Jiang-Huai (Yangtze River-Huaihe River), Central China, South China, East China, and Xinjiang regions, but showing relatively lower skill in Northeast China, Inner Mongolia, the Qinghai-Xizang Plateau, and Southwest China. Temporally, the Fengshun Model demonstrates superior prediction skill with 1 pentad and 4-8 pentad lead times, with the highest skill appearing at a 6 pentad lead time, and improvements in TCC, ACC and IPS can reach 42%, 260%, and 4.5%, respectively, which indicates an extended predictability window. This advantage is primarily attributed to the Fengshun Model’s better characterization of 500 hPa geopotential height anomalies over key circulation regions in mid-latitudes of Asia. However, the Fengshun Model shows a relatively weak prediction skill with 2 pentad to 3 pentad lead times, which is likely due to the decay of initial atmospheric signals and insufficient influence from underlying surface information. Future improvement focus will be put on incorporating multi-layer surface information to further enhance the Model’s prediction performance.