The effort to build computer models of the earth’s climate is nearly a half-century old. Models of global atmospheric circulation, and of atmosphere-ocean-ice interaction are by now immensely complex, but their conceptual structure is anchored in a simple “toy models” with very few parameters, containing most of the physics driving their much larger descendents. At the other end of the spectrum from “toy models” are Model Intercomparison Projects (MIPS) in which competing global models are assessed by running the identical data on different models and comparing outputs. The results of these MIPS are sobering. None of the major climate models accurately simulates the ocean-atmosphere system, and all must be doctored to make them work, whether by uncoupling the ocean and atmosphere, or by “flux adjustments,” the technical name for unphysical numerical corrections that bring model outputs in line with observed climate behavior. As interesting is the knowledge that these very large global climate models are not instruments of discovery: they do not tell us anything physical that we did not already know from the simple “toy” models of the 1950s-1970s. This paper explores the forces driving the evolution of these climate models, looks at the current impasse, and notes some recent, radical strategies to make the models better by making them smaller and simpler.