The recent advancements in robotic manipulation have significantly shifted towards integrating diffusion models and large language models to enhance dexterity, adaptability, and task-specific functionality. Diffusion models are being leveraged to streamline grasp synthesis and policy generation, offering faster inference and higher diversity in generated poses. Notably, these models are being adapted for hybrid frameworks that handle both discrete and continuous action spaces, improving exploration and policy diversity in reinforcement learning tasks. Additionally, the incorporation of large language models with quality diversity algorithms is enabling task-aware grasping, where semantic understanding and geometric reasoning are combined to select grasps based on specific tasks. This approach not only enhances the robot's ability to perform diverse tasks but also improves the transferability of learned skills to real-world scenarios. Furthermore, there is a growing focus on functional grasping for dexterous hands, where systems are being developed to enable one-shot transfer of human grasping poses to various robotic hands, facilitating robust sim-to-real transfer. The integration of multi-modal soft grippers with in-hand manipulation capabilities is also advancing, providing more versatile and general-purpose manipulation solutions. Overall, the field is progressing towards more intelligent, adaptable, and functionally diverse robotic manipulation systems.