​The subtropical gyres occupy about 70% of the ocean surface. While primary production (PP) within these oligotrophic regions is relatively low, their extension makes their total contribution to ocean productivity significant. Monitoring marine pelagic primary production across broad spatial scales, particularly across the subtropical gyre regions, is challenging but essential to evaluate the oceanic carbon budget. PP in the ocean can be derived from remote sensing however in situ depth-integrated PP (IPP^is) measurements required for validation are scarce from the subtropical gyres. In this study, we collected >120 IPP^is measurements from both northern and southern subtropical gyres that we compared to commonly used primary productivity models (the Vertically Generalized Production Model, VGPM and six variants; the Eppley-Square-Root model, ESQRT; the Howard–Yoder–Ryan model, HYR; the model of MARRA, MARRA; and the Carbon-based Production Model, CbPM) to predict remote PP (PP_r) in the subtropical regions and explored possibilities for improving PP prediction. Our results showed that satellite-derived PP (IPP^sat) estimates obtained from the VGPM1, MARRA and ESQRT provided closer values to the IPP^is (i.e., the difference between the mean of the IPP^sat and IPP^is was closer to 0; |Bias| ~ 0.09). Model performance varied due to differences in satellite predictions of in situ parameters such as chlorophyll a (chl-a) concentration or the optimal assimilation efficiency of the productivity profile (P^B_opt) in the subtropical region. In general, model performance was better for areas showing higher IPP^is, highlighting the challenge of PP prediction in the most oligotrophic areas (i.e. PP < 300 mg C m⁻² d⁻¹). The use of in situ chl-a data, and P^B_opt as a function of sea surface temperature (SST) and the mixed layer depth (MLD) from gliders and floats in PP_r models would improve their IPP predictions considerably in oligotrophic oceanic regions such as the subtropical gyres where MLD is relatively low (<60 m) and cloudiness may bias satellite input data.