Masked Diffusion Language Models (MDLMs) promise parallel token generation and arbitrary-order decoding, yet it remains unclear to what extent current models truly realize these capabilities. We characterize MDLM behavior along two dimensions -- parallelism strength and generation order -- using Average Finalization Parallelism (AFP) and Kendall's tau. We evaluate eight mainstream MDLMs (up to 100B parameters) on 58 benchmarks spanning knowledge, reasoning, and programming. The results show that MDLMs still lag behind comparably sized autoregressive models, mainly because parallel probabilistic modeling weakens inter-token dependencies. Meanwhile, MDLMs exhibit adaptive decoding behavior: their parallelism and generation order vary significantly with the task domain, the stage of reasoning, and whether the output is correct. On tasks that require "backward information" (e.g., Sudoku), MDLMs adopt a solution order that tends to fill easier Sudoku blanks first, highlighting their advantages. Finally, we provide theoretical motivation and design insights supporting a Generate-then-Edit paradigm, which mitigates dependency loss while retaining the efficiency of parallel decoding.

翻译：掩码扩散语言模型（MDLMs）承诺实现并行化词元生成与任意顺序解码，然而当前模型在多大程度上真正实现了这些能力尚不明确。我们从两个维度——并行强度与生成顺序——对MDLM的行为进行刻画，分别使用平均完成并行度（AFP）与肯德尔tau系数作为度量。我们在涵盖知识、推理与编程的58个基准测试上评估了八个主流MDLM（参数量最高达100B）。结果表明，MDLM在性能上仍落后于同等规模的回归模型，这主要源于并行概率建模削弱了词元间的依赖关系。同时，MDLM表现出自适应的解码行为：其并行度与生成顺序随任务领域、推理阶段以及输出正确与否而发生显著变化。在需要“逆向信息”的任务（如数独）上，MDLM倾向于采用一种先填充较简单空格的解题顺序，凸显了其优势。最后，我们提供了理论动机与设计思路，支持一种“先生成后编辑”的范式，该范式能在保留并行解码效率的同时缓解依赖关系的损失。