The PHC, ECCO2, SODA, GECCO3 and CMIP6 data were used to analyze the horizontal distribution characteristics, seasonal variation and long-term trend of the Arctic Ocean heat content, and analyze the simulation ability of the CMIP6 models in this paper. The results show that the heat content of the Arctic Ocean shows obvious seasonal change, with the lowest in April and the highest in September. Under historical circumstances (1850−2014), compared with the observation and reanalysis data, the heat content of the upper 500 m of the CMIP6 models ensemble average (MME) is warmer in the Greenland Sea, colder in the Norwegian sea, Barents Sea and Eurasian Basin, while the whole water column heat content of MME is warmer in almost all regions of the Arctic Ocean, with the largest deviation in the Greenland Sea. CMIP6 models have a large deviation in the simulation of Arctic Ocean temperature profile, and the average temperature of MME is higher than the observation and reanalysis data at the depth of more than 1 000 m. In the future case (2015−2100), the simulation of ocean heat content of MME shows obvious Arctic Ocean warming, but most of the Chinese models show no obvious warming situation. BCC-CSM2-MR and BCC-ESM1 are poor in simulating the annual mean heat content of the Arctic Ocean, CIESM is poor in simulating the seasonal and interdecadal variations of ocean heat content, while FIO-ESM-2-0 is good in simulating the annual heat content of the upper 500 m, the seasonal and interdecadal variations of heat content of the Arctic Ocean.