We study the debated contribution from thermally pulsing asymptotic-giant-branch (TP-AGB) stars in evolutionary population synthesis models. We investigate the spectral energy distributions (SEDs) of a sample of 51 spectroscopically confirmed, high-z (1.3 < zspec < 2.7), galaxies using three evolutionary population synthesis models with strong, mild and light TP-AGB. Our sample is the largest of spectroscopically confirmed galaxies on which such models are tested so far. Galaxies were selected as passive, but we model them using a variety of star formation histories in order not to be dependent on this pre-selection. We find that the observed SEDs are best fitted with a significant contribution of TP-AGB stars or with substantial dust attenuation. Without including reddening, TP-AGB-strong models perform better and deliver solutions consistent within 1σ from the best-fitted ones in the vast majority of cases. Including reddening, all models perform similarly. Using independent constraints from observations in the mid- and far-IR, we show that low/negligible dust attenuation, i.e. E(B - V) ≲ 0.05, should be preferred for the SEDs of passively selected galaxies. Given that TP-AGB-light models give systematically older ages for passive galaxies, we suggest number counts of passive galaxies at higher redshifts as a further test to discriminate among stellar population models.