A differentiated design of the constant stress accelerated degradation test is proposed to assess the lifetime of highly reliable and long lifetime products. The nonlinear Wiener process model with random effects is used to describe the degradation paths of the products. The maximum likelihood estimation is applied to estimate the unknown parameters of the model. Then, under the constraint of the total experimental budget, the differentiated plans of constant stress accelerated degradation tests are optimized by minimizing the asymptotic variance of the estimated mean time to failure. A genetic algorithm was developed to determine the number of units, inspection interval, and measurement times for the corresponding test of each stress level. Finally, an example and simulation experiments are presented to illustrate the effectiveness of the proposed method.