Gross primary productivity (GPP) represents the total carbon fixed by plants via photosynthesis, which is a fundamental component of the terrestrial carbon cycle. Yet its accurate estimation remains highly uncertain in regions with strong climate variability and ecosystem heterogeneity. In particular, monsoon-driven ecosystems, such as those across India, represent one of the most challenging and under-evaluated environments for global GPP products due to pronounced hydroclimatic variability, diverse vegetation types, and intensive land management practices, which introduce significant uncertainties in regional and global carbon budgets. Here, we conduct the first systematic and independent benchmarking of 11 widely used global GPP datasets using eddy covariance (EC) measurements spanning 14 sites representing the major land cover types in India, namely croplands, forests, grasslands, and mangroves. These datasets are generated using diverse methods, including remote sensing, process-based and light-use efficiency models, and machine learning (ML) algorithms. By integrating multi-metric statistical evaluation, Taylor diagram analysis, and hierarchical clustering, we assess the capability of these datasets and rank them to capture site-level variability in GPP. Our results reveal substantial ecosystem-specific discrepancies among these products. ML-based datasets (e.g., FLUXCOM-RS and XBASE) consistently perform well in croplands and forests, whereas process-based and microwave-derived products exhibit variable or degraded performance, particularly in grasslands. Notably, all datasets show reduced skill in mangrove ecosystems, highlighting fundamental limitations in representing the complex hydrological and biophysical processes. These findings indicate that global GPP products, although widely used, exhibit systematic biases under monsoon regimes and management-intensive landscapes. Our study demonstrates that monsoon-driven ecosystems provide a critical testbed for evaluating global carbon cycle products and reveals key limitations in current GPP estimation approaches. The results provide guidance for selecting appropriate datasets for regional applications and underscore the need for improved representation of hydrological variability and ecosystem complexity in next-generation GPP products and informed decision-making for carbon management and climate mitigation.