Understanding the 3D genome architecture is crucial for deciphering gene regulation in plants. However, most deep learning models are tailored for mammalian systems and fail to adapt to the diverse genomic landscapes of plants. To address this, we present IncepFold, a parameter-efficient, multimodal framework that integrates DNA sequence with chromatin accessibility (ATAC-seq) and H3K4me3 profiles via an Inception-based encoder. We validated IncepFold on four diverse plant species:Gossypium hirsutum, Zea mays, Solanum lycopersicum, and Sorghum bicolor. Across all four genomes, it significantly outperforms state-of-the-art baselines, including the transformer-based C.Origami, while using fewer parameters. Our input ablation studies confirm the critical contribution of each modality, and an in silico perturbation analysis demonstrates the model’s ability to identify functionally impactful genomic regions. Collectively, IncepFold provides a versatile, accurate, and computationally efficient framework for predicting 3D genome architecture across the plant kingdom, accelerating plant functional genomics and crop improvement research. The source code is available at: https://github.com/Yaonan789/IncepFold .