Cratonic lithospheric mantle is believed to have been formed in the Archean, but kimberlite-hosted coarse peridotites from Udachnaya in the central Siberian craton typically yield Paleoproterozoic Re-depletion Os isotope ages (TRD). By comparison, olivine megacrysts from Udachnaya, sometimes called “megacrystalline peridotites”, often yield Archean TRD ages, but the nature of these rare materials remains enigmatic. We provide whole-rock (WR) Re-Os isotope and PGE analyses for 24 olivine-rich xenoliths from Udachnaya as well as modal and petrographic data, WR and mineral major and trace element compositions. The samples were selected based on (a) high olivine abundances in hand specimens and (b) sufficient freshness and size to yield representative WR powders. They comprise medium- to coarse-grained (olivine < 1 cm) dunites, a megacrystalline (olivine > 1 cm) dunite, olivine megacrysts and low-orthopyroxene (11–21% opx) harzburgites equilibrated at 783–1154 °C and 3.9–6.5 GPa; coarse dunites have not been previously reported from Udachnaya; two xenoliths contain ilmenite. The harzburgites and dunites have similar WR variation ranges of Ca, Al, Fe, Cr and Mg# (0.917–0.934) typical of refractory cratonic peridotites, but the dunites tend to have higher MgO, NiO and Mg/Si. Mineral abundances and those of Ca and Al are not correlated with Mg#WR; they are not due to differences in melting degrees but are linked to metasomatism. Several samples with high 187Re/188Os show a positive linear correlation with 187Os/188Os with an apparent age of 0.37 Ga, same as eruption age of host kimberlite. Robust TRD ages were obtained for 16 xenoliths with low 187Re/188Os (0.02–0.13). TRD ages for low-opx harzburgites (1.9–2.1 Ga; average 2.0 ± 0.1 Ga, 1 σ) are manifestly lower than for dunites and megacrysts (2.4–3.1 Ga); the latter define two subsets with average TRD of 2.6 ± 0.1 Ga and 3.0 ± 0.1 Ga, and TMA of 3.0 ± 0.2 Ga and 3.3 ± 0.1 Ga, respectively. Differences in olivine grain size (coarse vs. megacrystalline) are not related to age. The age relations suggest that the dunites and megacrysts could not be produced by re-melting of harzburgites, e.g. in arc settings, nor be melt channel materials in harzburgites. Instead, they are relict fragments of lithospheric mantle formed in the Archean (likely in two events at or after 2.6 Ga and 3.0 Ga) that were incorporated into cratonic lithosphere during the final assembly of the Siberian craton in the Paleoproterozoic. A multi-stage formation of the Siberian lithospheric mantle is consistent with crustal basement ages from U-Pb dating of zircons from crustal xenoliths at Udachnaya and detrital zircons from the northern Siberian craton (1.8–2.0, 2.4–2.8 and 3.0–3.4 Ga). The new data from the Siberian and other cratons suggest that the formation of strongly melt-depleted cratonic lithosphere (e.g. Mg# ≥0.92) did not stop at the Archean-Proterozoic boundary as is commonly thought, but continued in the Paleoproterozoic. The same may be valid for the transition from the ‘Archean’ (4–2.5 Ga) to modern tectonic regimes.