@polvani - I’m also interested in this project. I was curious about this a couple weeks ago and plotted the global mean tas trends. With my set of models (13; trying to compare across generations, e.g., CAM5-CAM6) it appears that the multimodel mean historical trend went down in CMIP6 in the subset of models I looked at (historical / rcp8.5 in CMIP5; historical in CMIP6). This is pretty interesting, right?
Note there is a connection to Sample Model Like Observations project (I was interested in sampling surface trends like observations; this could be extended to other fields, too (as noted in that thread).
edit to note that “it went down” might be better described as it is basically the same, despite the widespread attention over ECS trends going up in CMIP6.