It is true that the model with the overall best performance is on the Pareto frontier, but it is not necessarily true that a given model's best cost-performance tradeoff is at max reasoning. (IE you could have a model that is on the frontier at medium reasoning, but at max reasoning becomes more expensive and not sufficiently good enough to remain on the frontier.)