We explore the robustness of the calibration of stellar models achievable with Ai Phe binary system. By means of the SCEPtER pipeline, we investigated the impact of different assumptions about the surface efficiency of microscopic diffusion. In the reference scenario, we allowed modification of the surface metallicity due to microscopic diffusion, while in the alternative scenario we assumed that competing mixing from other sources cancels out this effect. Due to the fact that the primary star has already experienced the first dredge-up while the secondary has not, the tested scenarios show interesting differences. While the estimated age is quite robust ($4.70^{+0.13}_{-0.14}$ Gyr and $4.62^{+0.13}_{-0.06}$ Gyr), the calibration of the convective core overshooting parameter $\beta$ reveals noticeable differences. The reference scenario suggests a wide multi-modal range of possible values of $\beta$ around 0.10; the alternative scenario computations point towards a sharp and lower $\beta$, around 0.04. The impossibility to obtain an unambiguous fit confirms the difficulty in achieving a sensible calibration of the free parameters of stellar models using binary systems, even when very accurate masses and radii are available.