Gavin Schmidt is the second author of that paper: even if that statement conflates different model versions or forcing measures I don't think I can be blamed for relying on it. (climateaudit.org)
For example, contrasting the development improvements or setbacks from different model versions in relation to the distribution of structural errors in the CMIP multimodel ensemble can provide an objective assessment as to whether model performance changes are significant. (eos.org)
He conflates different model versions (fully interactive simulations in Shindell et al (the p3 runs in CMIP5), with the non-interactive runs used in MEA15 (p1 runs)-RRB-, and different forcing definitions (Fi and Fa). (realclimate.org)