Closed vs open model gap may be smaller than assumed

Reddit post argues closed model providers may infer additional compute and tricks beyond model inference, skewing benchmark comparisons. Uses Claude vs GLM-5.2 to illustrate that superior scores don't necessarily reflect better architecture.