Abstract Accurate quantification of terrestrial gross primary production (GPP) is integral for enhancing our understanding of the global carbon budget and climate change. The light use efficiency (LUE) model is undoubtedly the most extensively applied method for GPP estimation. However, the two-leaf (TL)-LUE model using a ‘potential’ sunlit leaf area index (LAI_su) can separate a portion of LAI_su even when the canopy does not receive any direct radiation, leading to the underestimation of GPP under cloudy and overcast days. Here, we developed a dynamic-leaf (DL) LUE model by introducing an ‘effective’ LAI_su to improve GPP estimation, which considers the comprehensive contribution of LAI_su when the canopy does and does not receive direct radiation. In particular, the new model decreases LAI_su to zero when direct radiation reaches zero. Our evaluation at eight ChinaFLUX sites showed that (1) the DL-LUE model outperformed the most well-known BL-LUE (namely, the MOD17 GPP algorithm) and TL-LUE models in reproducing the daily in situ GPP, especially at four forest sites [reducing the root mean square error (RMSE) from 1.74 g C m⁻² d⁻¹ and 1.53 g C m⁻² d⁻¹ to 1.36 g C m⁻² d⁻¹ and increasing the coefficient of determination (R ²) from 0.74 and 0.79–0.82, respectively]. Moreover, the improvements were particularly pronounced at longer temporal scales, as indicated by the RMSE decreasing from 29.32 g C m⁻² month⁻¹ and28.11 g C m⁻² month⁻¹ to 25.81 g C m⁻² month⁻¹ at a monthly scale and from 231.82 g C m⁻² yr⁻¹ and 221.60 g C m⁻² yr⁻¹–200.00 g C m⁻² yr⁻¹ at a yearly scale; (2) the DL-LUE model mitigated the systematic underestimation of the in situ GPP by both the TL-LUE and BL-LUE models when the clearness index (CI) was below 0.5, as indicated by the Bias reductions of 0.25 g C m⁻² d⁻¹ and 0.46 g C m⁻² d⁻¹, respectively; and (3) the contributions of the shaded GPP to the total GPP from the DL-LUE model were higher by 0.07–0.16 than those from the TL-LUE model across the eight ChinaFLUX sites. The proposed parsimonious and effective DL-LUE model not only has great potential for improving global GPP estimations but also provides a more mechanism-based approach for partitioning the total GPP into its shaded and sunlit components.