Unified monocular 3D object detection is crucial for robotics and autonomous driving, aiming to develop robust models that can effectively perform across diverse scenes using a single image. Recent methods, such as CubeRCNN and UniMODE, have achieved state-of-the-art (SOTA) performance on benchmarks like Omni3D using CNN or ViT backbones. However, these methods often face computational efficiency challenges, particularly as they rely on complex self-attention mechanisms in ViTs. In contrast, State Space Models (SSMs), especially Mamba variants like EfficientVMamba, offer an effective alternative to the \(O(N^2)\) self-attention in ViTs with their linear \(O(N)\) complexity. This paper introduces MambaMODE, a novel unified monocular 3D detector that integrates the efficient EfficientVMamba backbone into UniMODE’s two-stage framework. The proposed model effectively combines the strengths of both architectures, offering enhanced computational efficiency. Extensive experiments on the Omni3D dataset show that MambaMODE achieves 3D detection performance comparable to SOTA methods such as UniMODE. Importantly, MambaMODE significantly reduces model parameters and computational overhead while improving training speed by 60%. These efficiency improvements highlight MambaMODE’s applicability on resource-constrained platforms and its substantial potential for advancing the unified 3D object detection task.