We analyze fluctuations in a "hybrid" atomistic model mimicking CO oxidation on nanoscale facets of metal(100) catalyst surfaces. The model incorporates a mean-field-like treatment of infinitely mobile CO, and a lattice-gas treatment of the superlattice ordering of immobile O. For an infinite system, it exhibits an Ising-type order-disorder transition for O, together with mean-field-like bistability disappearing at a cusp bifurcation. For finite systems, we use kinetic Monte Carlo simulation to study the probability distribution for the population of adsorbed species, from which bistability can be observed, together with fluctuation-induced transitions between the two stable states. An effective potential picture emerges from our analyses that can be used to quantify both the system size dependence of fluctuations and the transition rates. Thus, our hybrid atomistic model displays fluctuation behavior analogous to traditional mean-field models. This qualitative behavior can be understood by approximate treatments of population dynamics using master equations and Fokker-Planck equations. A generalized model with finite mobility of CO is also analyzed for comparison with the hybrid model. In contrast, it exhibits fluctuation behavior akin to equilibrium systems with Ising-type first-order transitions.