Face recognition technology has advanced significantly with the development of deep learning architectures and the availability of large-scale training datasets. This paper presents a comprehensive review and benchmarking of face recognition models, from early classical architectures to the latest state-of-the-art techniques. The manuscript meticulously utilizes a broad spectrum of deep neural networks: VGG-16, FaceNet, DeepFace, ArcFace, MobileFaceNet, MagFace, SFace, and transformer-based approaches, and rigorously evaluates benchmark datasets such as LFW, AgeDB-30, CelebA+masks, and CASIA-WebFace. This paper presents a comparative analysis of state-of-art face recognition models, assessing their accuracy, robustness, model size, deployment efficiency and inference speed for real-time industrial applications, where RetinaFace demonstrated superior performance across all models, establishing itself as the optimal choice for real-world scenarios.