Please find my comments in the attachment.

This manuscript presented a thorough and fundamental evaluation of the CMIP6 model simulations of PM2.5 over China. CMIP6 simulations are widely applied for climate related studies and it is necessary to fully understand model uncertainty as aerosol plays an important role in the climate system. Yet the performance of global model for predicting surface PM2.5 concentrations has been largely ignored as it was considered as a special task for chemical transport models, especially at regional scale. However, in recent decade many studies reported the significance of interactions between air pollution and climate, thus it is important to reveal how the CMIP6 simulations can reproduce surface PM2.5 as well. The manuscript is well organized with clear description of model and observational data employed. It provide a thorough discussion of the results and origins of uncertainties with solid method. Therefore, I would recommend it to be accepted with minor revisions if the following comments could be properly addressed.

line158: It would be helpful to show the comparison between direct summary of all fine aerosol species (e.g., PM2.5 = sulfate + oa + nitrate + ammonium + bc + fine dust + fine sslt + bc) and the value of this equation, for the 4 models which provide nitrate. This would help to demonstrate the accuracy of the equation.

line171: It would be helpful to show comparison between satellite product and model results for AOD at monthly scale to reveal the performance of model in simulating seasonal variations of aerosol over China. This may provide more indications of model uncertainty, such as dust may dominate in spring and OA may dominate in summer.

line194: Not sure what is “effect of interannual variability”, please make it clear.

line201: A figure similar to Fig.4 but for bias between model and satellite-based product would be helpful to reveal the difference more clear.

line209: It would be helpful to provide a brief discussion of why certain specific model show better performance than others.

line238: Can you add a brief discussion of possible causes for the decline over 200-2005 in satellite data? Is it an observational fact or satellite bias, if it is a fact, then why model can not reproduce it?

line250: Not sure what is “spatially coincidently sampled”, please make it clear.

line255: It would be helpful to add a brief discussion to explain why model difference peaks at these regions.

line259: Is it because of spatial distribution pattern of CEDS emission?

line269: How much was CEDS emission data, as compared to which dataset?

line272: BC is primary

line272: Do all models have warm bias over Xinjiang? In addition, some observation sites provide PBL measurements as well, so why not perform evaluation of simulated PBL directly?

line292: It would be helpful to briefly explain how 14 models simulate sulfate formation chemistry.

line344: Not sure what is “seasonal model results to match seasonal observational”, do you mean conduct model evaluation at seasonal scale, so all models and observations are averaged seasonally?

line406: I would recommend to mention it as those causes for aerosol underestimation may also affect O3.