Artificial photosynthesis (AP) promises to replace society's dependence on fossil energy resources via conversion of sunlight into sustainable, carbon-neutral fuels. However, large-scale AP implementation remains impeded by a dearth of cheap, efficient catalysts for the oxygen evolution reaction (OER). Cobalt oxide materials can catalyze the OER and are potentially scalable due to the abundance of cobalt in the Earth's crust; unfortunately, the activity of these materials is insufficient for practical AP implementation. Attempts to improve cobalt oxide's activity have been stymied by limited mechanistic understanding that stems from the inherent difficulty of characterizing structure and reactivity at surfaces of heterogeneous materials. While previous studies on cobalt oxide revealed the intermediacy of the unusual Co(IV) oxidation state, much remains unknown, including whether bridging or terminal oxo ligands form O2 and what the relevant oxidation states are. We have addressed these issues by employing a homogeneous model for cobalt oxide, the [Co(III)4] cubane (Co4O4(OAc)4py4, py = pyridine, OAc = acetate), that can be oxidized to the [Co(IV)Co(III)3] state. Upon addition of 1 equiv of sodium hydroxide, the [Co(III)4] cubane is regenerated with stoichiometric formation of O2. Oxygen isotopic labeling experiments demonstrate that the cubane core remains intact during this stoichiometric OER, implying that terminal oxo ligands are responsible for forming O2. The OER is also examined with stopped-flow UV-visible spectroscopy, and its kinetic behavior is modeled, to surprisingly reveal that O2 formation requires disproportionation of the [Co(IV)Co(III)3] state to generate an even higher oxidation state, formally [Co(V)Co(III)3] or [Co(IV)2Co(III)2]. The mechanistic understanding provided by these results should accelerate the development of OER catalysts leading to increasingly efficient AP systems.