They do "double substitution procedure" (another google term) for most masses for about three grand plus or minus incidentials (cleaning, shipping). Unfortunately the best descriptions I can find of NIST double substitution procedure is docx unfree format.
However it boils down to a really high quality double arm balance (like a lever with a pivot in middle and buckets on each end) and its temp controlled to keep the lengths of the arms identical and they stick your (test) mass on one side, stack calibration weights on the other side and record what balances, then remove your mass and stack more calibration masses in its place optimistically very similar in mass to the other side, but it won't be equal because one arm is probably a nanometer longer than the other arm or something. Essentially they're testing it twice in order to factor out a difference in arm lengths however small. You can get around eight or nine sig figs of accuracy now a days using this procedure.
I would enjoy seeing an electrostatic force balance like they use for microgram scale masses.