There is no model that consistently yields either highest or lowest warming because each model has strengths and weaknesses which depend upon the conceptual emphases. This varies with latitude, continentality, proximity of water/lakes, presence/absence of ice and mountains, regional significance of atmospheric processes - both physical and chemical, regional airflow strengths such as monsoonal and hurricanes, etc. Each model calibration works either brilliantly well, or downright dubiously, according to the model conceptualization. There is no 'one size fits all' (environments).
That said, however, I routinely use the CMIP-5 ensemble as listed in the World Bank Climate Change Knowledge Portal - just because it is convenient and readily available. You can find it at http://sdwebx.worldbank.org/climateportal/index.cfm - just click on the world map for your region / country / area of interest, and select the parameters you want. This is not the full ensemble, but 16 selected models that seem to work well - at least most of the time.
So here is a recent example from Nepal: scanning the maximum and minimum monthly temperatures, 'giss-r' generally yielded the hottest temperatures, although for occasional months the hottest was 'ccsm-4'. The coolest temperatures were mostly given by 'noresm 1-m' though for occasional months the coldest was from 'miroc esm'. However, this is just a random example. I do not believe that such conclusions are consistent everywhere in the world.
Why concentrate on the coolest and hottest models? Doesn't it make more sense to work with the ensemble median? I find that the relative distribution of model results is roughly similar regardless of the RCP. I usually quote RCP 4.5 or 6.0. RCP 8.5 is too pessimistic. Does anyone still seriously think that RCP 2.8 is attainable?
My advice is to spend a day picking temperature trends from the WB-CMIP model portal for a wide range of environments, and only then draw your conclusions.