Accurate gross primary productivity (GPP) estimation in Tibetan Plateau alpine meadows is crucial for understanding carbon cycling under climate change. The performance of 14 diverse light use efficiency (LUE) models (including CASA, MOD17, MuSyQ, P-model, Wang-model) in simulating daily GPP against eddy covariance (EC) measurements from the Haibei site (2015—2020) was comprehensively evaluated. It was specifically investigated how differences in model structure, particularly the formulation of environmental stress functions for temperature (f(T)), water availability (f(W)), and radiation/sky conditions (e. g., cloudiness index, CI), influence simulation accuracy in this unique high-altitude environment characterized by low temperatures, intense radiation, and distinct hydrology. Performance metrics (R2, RMSE, MAE) revealed substantial variations. MuSyQ demonstrated the best performance (R2=0.59, RMSE was 0.89g/(m2·d)), followed by Wang-model and CI-LUE. Models explicitly incorporating sky conditions (CI) or differentiating diffuse/direct radiation impacts generally outperformed simpler models. Conversely, models like TEC and MOD17 showed poor results (R2≤0.25). The key factors influencing simulation accuracy lie in how models responded to the plateau's low temperatures, intense radiation, and special hydrological conditions. Specifically, the model's parameterization of low-temperature responses, including thresholds, optimum temperature, and the expression of photo inhibition effects under low-temperature and intense radiation, directly determined the accuracy of temperature stress simulation. The capability of water stress parameterizations to effectively capture the plateau's unique “moist soil-dry atmosphere” water limitation was key to accurate water stress simulation. Furthermore, the introduction of sky conditions and their regulation of LUE to reflect diffuse light advantages and strong light inhibition directly affected the characterization of radiation use efficiency. Regarding the water stress function, models employing the land surface water index (LSWI) performed relatively better among comparable models than those based on vapor pressure deficit (VPD), soil moisture, or the ratio of actual to potential evapotranspiration. Notably, no model achieved an R2 exceeding 0.6, highlighting a persistent challenge in accurately simulating GPP in this ecosystem. The results underscored the critical need for refining LUE models, particularly their environmental stress functions, to better capture the complex physiological responses of alpine vegetation to the specific conditions of the Tibetan Plateau, thereby improving regional GPP estimation.