Compared with the observations from 55 Beijing auto weather stations from 1 October 2018 to 30 September 2019, the European Center MediumRange Weather Forecasts Fine Grid Model (ECMWFthin), Regional Assimilation and Prediction System (Grapes), Rapidrefresh Multiscale Analysis and Prediction System - Short Term (RMAPSST), Central Station Guided Forecast (SCMOC), Beijing Intelligent Grid Temperature Objective Prediction Method (BJTM) and Analog Ensemble method (AnEn) which mainly focus on the daily maximum and minimum temperature in the Beijing area are evaluated. (1) In total, the results show that ECMWFthin model performance was better than Grapes and RMAPS; Two objective methods, BJTM and AnEn, had apparent improvement effects on ECMWFthin. (2) AnEn performed well from October 2018 to April 2019, and BJTM performed well from May to September 2019. Regarding different forecast timeliness, AnEn performed well in the shortterm and the first part of mediumterm, BJTM performed well in 5 to 9 days in mediumterm. (3) Focusing on the Guanxiangtai station, the systematic deviation was evident in all three models. Objective methods reduced the systematic deviation of models. (4) Under the background of precipitation, wind and no obvious weather, the two objective methods BJTM and AnEn had significantly improved the forecast quality of the ECMWFthin model for daily maximum temperature. However, when haze weather happened, the forecast accuracy of ECMWFthin was significantly higher than other models and methods. For the minimum daily temperature, except for precipitation weather background, the ECMWFthin model had the smallest deviation, and objective methods slightly improved the model results. Moreover, the RMAPS results showed better performance when precipitation occurred, and objective methods reduced the systematic deviation of the largescale model.