Right, de novo mutation rate is much lower (even lower than what you wrote, just slightly above 1e-8 according to deCODE). The 1/1000 rate is the average difference between two random individuals (not twins), which corresponds to your 3M SNPs per 3B bases.