Abstract: As neural networks integrate more with additive and subtractive hybrid manufacturing (ASHM), researchersurgently need high-quality large 3D model datasets to train and validate neural network-based ASHM process planning methods. To address issues like scarce, scattered existing applicable 3D models, existing editors’ failure to realize both local adjustment and rapid model generation, and extremely high dataset generation costs, a large-scale 3D model dataset DHMP for ASHM process planning is proposed. First, an interactive 3D model generation system GMHM is developed, which uses solid geometry construction to batch-generate 3D models based on the discrete mesh representation of primitive models. Second, a gap detection algorithm is applied to ensure models meet required geometric feature constraints—it uniformly samples potential tool directions on the Gaussian upper hemisphere and simulates tool motion in each direction to check for gaps. Finally, the DHMP dataset (over 100 000 models and partly with VASCO-based process planning annotations) is generated via GMHM. Experiments show GMHM’s generation efficiency is 1 558 times higher than AutoCAD and Rodin; DHMP outperforms Thingi10k and ABC in average curvature, surface area-to-volume ratio, and manufacturability (100%).