Diffuse Thinking: Exploring Diffusion Language Models as Efficient Thought Proposers for Reasoning

Large language models (LLMs) have demonstrated strong capabilities in complex reasoning tasks, yet the multi-step reasoning processes often lead to expensive computation cost.The recent advance of diffusion language models (DLMs) adopts a parallel, non-autoregressive generation mechanism, which enables the efficient production of multiple outputs.In this paper, we explore a collaborative reasoning framework that combines diffusion-based generation with autoregressive evaluation.Specifically, we leverage DLMs to efficiently propose stepwise reasoning thoughts, and employ LLMs as evaluators to assess and select candidates based on their plausibility and correctness.By decoupling proposal generation from evaluation, our framework exploits the strengths of both models: efficient exploration from diffusion models and causally grounded assessment from autoregressive models, which naturally aligns with the divergent-convergent thinking framework in cognitive psychology.Experiments across various mathematical and logical reasoning benchmarks demonstrate that, our framework improves inference efficiency while maintaining competitive or superior reasoning accuracy, laying the groundwork for building efficient reasoning architectures.Our code is open-source at https://anonymous.4open. science/r/