ABSTRACT We conduct the first 3D hydrodynamic simulations of oxygen–neon white dwarf–neutron star/black hole mergers (ONe WD–NS/BH mergers). Such mergers constitute a significant fraction, and may even dominate, the inspiral rates of all WD–NS binaries. We post-process our simulations to obtain the nuclear evolution of these systems and couple the results to a supernova spectral synthesis code to obtain the first light curves and spectra for these transients. We find that the amount of 56Ni synthesized in these mergers grows as a strong function of the WD mass, reaching typically 0.05 and up to $0.1\, {\rm M}_\odot$ per merger. Photodisintegration leads to similar amounts of 4He and about a ten times smaller amount of 1H. The nuclear yields from these mergers, in particular those of 55Mn, may contribute significantly to Galactic chemical evolution. The transients expected from ONe WD–NS mergers are dominantly red/infrared, evolve on month-long time-scales and reach bolometric magnitudes of up to −16.5. The current surveys must have already detected these transients or are, alternatively, putting strong constraints on merger scenarios. The properties of the expected transients from WD–NS mergers best agree with faint type Iax supernovae. The Vera Rubin Observatory (LSST) will be detecting up to thousands of merging ONe WD–NS systems per year. We simulate a subset of our models with 2D axisymmetric flash code to investigate why they have been challenging for previous studies. We find that the likely main challenge has been effectively modelling the nuclear statistical equilibrium regime in such mergers.