It is difficult to achieve low expense and high accuracy in computational methods, yet it remains a key objective in atomistic approaches. In solid state physics, advanced atomistic potentials using reactive force fields have shown promise in delivering both. However, these methods have not been applied widely beyond their development environment and thus their strengths and weaknesses are not fully understood. In this work we present benchmark calculations on silica (SiO2) and hafnia (HfO2) structures, comparing a leading charge optimized many-body potential to a more advanced density functional calculation. We find that although the atomistic potential gives excellent results for bulk structures, it has severe shortcomings when applied to small systems with low coordinated atoms. We also establish clearly the components of the many-body potential and how these relate to predicted physical properties.