Chlorophyll (Chl) is widely taken as a proxy for phytoplankton biomass, despite well known variations in Chl:biomass ratios as an acclimative response to changing environmental conditions. For the sake of simplicity and computational efficiency, many large scale biogeochemical models ignore this flexibility, compromising their ability to capture phytoplankton dynamics. Here we evaluate modelling approaches of differing complexity for phytoplankton growth response: fixed stoichiometry, classical variable-composition with photo-acclimation, and Instantaneous Acclimation with optimal resource allocation. We compare the performance of these models against biogeochemical observations from time-series sites BATS and ALOHA, where phytoplankton composition varies substantially. Models including photo-acclimation capture the observations better with minimal parameter tuning and are more portable. Compared to the classical variable composition approach, instantaneous acclimation yields similar performance and portability, while requiring fewer state variables. Further assessments using objective optimisation and more contrasting stations are suggested.