Window glass is a ternary mixture, while pyrex (after window glass, the most common form of commercial glass) is a quaternary. Building on our previous success in deriving the composition of window glass (sodium calcium silicate) without adjustable parameters, and borrowing from known reconstructed crystalline surfaces, we model pyrex as silica clusters with a specific ternary interface. Our global model explains the thermal expansivity contours of ternary sodium borosilicates, and it is consistent with the optimized resistance of pyrex to mechanical and thermal shocks.It suggests new directions for studying the nanoscopic structure of these remarkable materials.In principle the structure of glasses is an exponentially complex combinatorial problem.In the early days of glass science it was customary to dismiss this problem by saying that network glasses, for example, form "continuous random networks", or that metallic glasses could be modeled by "random" packing of hard spheres. Modern glass theory goes far beyond these early models, but there are still many approaches to understanding glass structure and properties. One approach, that has the advantage of being close to commercial practice, is variational (standard for mathematical studies of non-polynomial complete (complex) problems) and focuses attention on optimizing structural properties as a function of composition. Here we adopt this approach to study pyrex (next to