A longtime project of ours was just published as http://dx.doi.org/10.1039/C6CP00688D in PCCP.
We revise the energetics of the S66x8 noncovalent interactions benchmark by means of explicitly correlated coupled cluster methods. Using the new reference data, we carry out an extensive benchmark of DFT and lower-level ab initio methods. DSD-PBEP86-NL turns out to have an accuracy rivaling ab initio methods. The advantage of double hybrids or RPA-based correlation is shown to be in the medium distance range: even dRPA75 still benefits from a dispersion correction.
We also propose a novel index for the nature of noncovalent interactions (dispersion-dominated vs. electrostatic-dominated) based on
A number of ancillary methodological issues are addressed in passing.