The addition of solar forcing to CMIP6 means that they completely missed it for the first 5 models, so the consensus was wrong. Science is driven by observation and experiment not by consensus.
We should not expect their first iteration to accurately model clouds, but at least they are factoring it in now. How much did consensus override the concerns about cloud cover during the formation of climate models over the last few decades? Consensus is just a fancy word for groupthink.
We should be asking how much more accurate do our models need to be in order to predict future climate?
There are cycles such as the Atlantic Multi-decadal Oscillation that we still don't really know how to model with a high degree of accuracy. The AMO has dramatic effects on climate. So does the PDO.
We don't know how volcanic activity is going to play out over the 21st century.
We don't know how much cosmic ray flux will increase if we enter a prolonged solar minimum cycle like the Maunder Minimum.
There are a ton of factors that influence climate that our models don't account for.