Recent advances in foundation models, large language models, and generative models are beginning to reshape how researchers approach robot co-design. At the same time, robot co-design has long relied on iterative paradigms such as reinforcement learning, optimization, and evolutionary search. While both directions have shown promise, their relative strengths, limitations, and opportunities for integration remain unclear. This workshop focuses on learning-based robot co-design, with particular emphasis on two complementary paradigms: generative approaches that propose novel robot embodiments and morphologies, and iterative approaches that improve designs through reinforcement learning, optimization, or evolutionary search. By bringing these perspectives together, the workshop aims to foster discussion on how recent advances in generative AI are changing the co-design landscape, when different paradigms are most effective, and whether hybrid approaches can unlock new capabilities. The workshop will bring together researchers working on robot co-design, generative design, reinforcement learning, and optimization-based design, providing a platform to present recent advances and discuss future directions for learning-based robot co-design.