一句话：首次系统评估扩散大语言模型在代码生成、缺陷检测与程序修复中的表现，发现其准确率与推理延迟均优于自回归模型。
自回归大语言模型在软件工程中应用广泛，但难以处理代码结构信息且推理延迟高。扩散大语言模型（DLLMs）具备全局双向编码与解耦生成步骤，有望克服上述限制。本文首次对 DLLMs 在软件开发生命周期中的表现进行全面评估，涵盖代码生成、缺陷检测与程序修复。作者在包含 52,937 个任务的大规模基准上测试 7B 参数 DLLMs，并与自回归模型对比。结果显示 DLLMs 平均准确率提升 30%，跨文件修复任务上提升 113%，同时保持更优效率与更低延迟。该工作确立了 DLLMs 作为软件工程任务的更优范式，并公开了代码与数据集。
Autoregressive Large Language Models (AR-LLMs) are widely used in software engineering (SE) but face limitations in processing code structure information and suffer from high inference latency. Diffusion LLMs (DLLMs) offer a promising alternative with global bidirectional encoding and decoupled generation steps. This work presents the first comprehensive evaluation of DLLMs across the software development lifecycle, including code generation, defect detection, and program repair. On a large-scale benchmark of 52,937 tasks, 7B-parameter DLLMs outperform AR-LLMs with a 30% average accuracy improvement, achieving a 113% gain on cross-file repair, while maintaining superior efficiency and reduced latency. Our results establish DLLMs as a superior paradigm for SE tasks. The code is publicly available at https://anonymous.4open.science/r/all_datasets-B4A6
@inproceedings{paperbot12029,
title = {Unveiling the Potential of Diffusion Large Language Models in Software Engineering Tasks: An Empirical Study},
author = {Jingyao Zhang and Tianlin Li and Xiaoyu Zhang and Qiang Hu and Bin Shi},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3786582.3786842}
}