The field of molecular design is witnessing a significant shift towards the use of generative models, which have shown great promise in generating diverse and high-reward structured objects. These models are being applied to various tasks such as drug discovery, peptide design, and biomolecule generation, and are demonstrating improved performance over traditional methods. A key trend in this area is the development of discrete flow generative models, which are able to learn complex distributions over discrete spaces and generate high-quality samples. Another important direction is the integration of multi-objective guidance into generative frameworks, allowing for the simultaneous optimization of multiple therapeutic properties. The use of geometry-aware frameworks and Riemannian geometry is also being explored, providing a powerful new paradigm for molecular design. Notable papers in this area include: Refine Drugs, Don't Complete Them, which introduces a discrete flow generative model for fragment-based drug discovery, and AReUReDi, which presents a discrete optimization algorithm with theoretical guarantees of convergence to the Pareto front. Additionally, MG2FlowNet and PepCompass demonstrate the effectiveness of enhanced sampling strategies and geometry-aware frameworks in generating high-reward samples and discovering novel molecules.