Abstract: Material discovery is a critical area of research with the potential to revolutionize various fields, including carbon capture, renewable energy, and electronics. However, the immense scale of the chemical space makes it challenging to explore all possible materials experimentally. To address this, we introduce FlowLLM, a novel generative model that combines large language models (LLMs) and conditional flow matching (CFM) to design novel crystalline materials. FlowLLM leverages the strengths of both models: the LLM learns an effective base distribution for the CFM model, generating an initial material representation that is iteratively refined using the CFM model. Our experiments demonstrate that FlowLLM significantly outperforms state-of-the-art methods, increasing the generation rate of stable materials by over $3\times$. This represents a substantial advancement in generative modeling for material discovery, with the potential to accelerate progress in material science.