The field of diffusion language models is moving towards more flexible and accurate text generation and reasoning capabilities. Recent developments have focused on introducing mechanisms for self-correction and refinement, allowing models to detect and revise low-quality tokens and generate more coherent text. Another key direction is the development of methods for guiding the reasoning process in diffusion language models, enabling them to solve complex problems more effectively. Notable papers in this area include: RFG: Test-Time Scaling for Diffusion Large Language Model Reasoning with Reward-Free Guidance, which proposes a principled method for guiding the reasoning trajectory of diffusion language models without explicit process reward. PRISM--Plug-in Remasking for Inference-time Self-correction of Masked Diffusions, which introduces a lightweight, model-agnostic approach for self-correction in masked diffusion models. Step-Aware Policy Optimization for Reasoning in Diffusion Large Language Models, which proposes a novel RL algorithm that aligns the diffusion language model's denoising process with the latent reasoning hierarchy.