I am working on GCMs's outputs & CMIP5. I can say: We can't say, what is the best model? You know it is very overall.Defiantly it depends on your case study and its climatology and the others characteristics of that case study, such as geographical adjectives. If you want to know the best model, I think that you should run all the models & after that comparing the results of them with observation data with a experience period to find the errors. May be it is better to filtering the models, for example you study the papers that these are the same characteristics of your study area and after that you can select the best models according to those papers.
I hope so, that I can explain my conception in a proper way.