Elamur Research⌕
☰
home / papers / 2026 / The Bitter Lesson of Diffusion Language Models for Agentic Workflows: A Comprehensive Reality Check

The Bitter Lesson of Diffusion Language Models for Agentic Workflows: A Comprehensive Reality Check

Paper · 2026 · Qingyu Lu, Liang Ding, Kanjian Zhang et al.

Summary

Diffusion-based language models (dLLMs) decode tokens in parallel and are often promoted as a fix for the slow, turn-by-turn latency of LLM agents. This paper asks whether that speed comes with usable agentic competence. The authors compare four dLLMs (Llada-8B, Dream-7B, FdLLM-7B, DVar-8B) against two autoregressive models (Qwen-8B, Ministral-8B) on two kinds of tasks:

  • Embodied tasks from AgentBoard, using a ReAct loop.
  • Tool-calling tasks from BFCL-v3.

The dLLMs perform far worse in both settings. In embodied environments they keep repeating the same failed action instead of trying a new plan. When calling tools, they often break the required JSON structure or invent API arguments.

To find where dLLMs might still help, the authors built DiffuAgent, a multi-agent evaluation framework. It plugs a dLLM into one auxiliary role inside an LLM-driven pipeline. In this setup, dLLMs work reasonably well in non-causal roles such as memory summarization, spotting redundant trajectories and picking relevant tools.

The work matters because it shows that faster decoding does not by itself produce reliable agents. Diffusion models need denoising that respects causal, precise reasoning before they can serve as agent backbones.

Key points

  • Benchmarks: AlfWorld (134 tasks), ScienceWorld (90) and BabyAI (112) via AgentBoard, scored by success and progress rate. BFCL-v3 was subsampled to at most 50 items per category, giving 758 examples.
  • Embodied results: average success was 45.0% for Qwen-8B and 31.8% for Ministral-8B. The dLLMs scored between 2.0% (DVar-8B) and 7.5% (Llada-8B).
  • Multi-turn BFCL: every dLLM scored 0.0% on all multi-turn BFCL categories. Qwen-8B reached an overall tool-calling score of 57.8%.
  • Speed: only FdLLM-7B (about 151 tok/s) and DVar-8B (about 167 tok/s) were faster than the autoregressive baselines. Llada-8B and Dream-7B were both slower and less accurate.
  • Scope: the study is inference-only, with no agent-specific fine-tuning or RL. It covers only AgentBoard's embodied subset and BFCL v1–v3.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Nanyang Technological UniversitySingapore · Singapore156 works · 25,644 citations🎓Southeast UniversityNanjing · China28 works · 354 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,320 works

Cite

Cite as Qingyu Lu, Liang Ding, Kanjian Zhang, Jinxia Zhang and Dacheng Tao. 2026. The Bitter Lesson of Diffusion Language Models for Agentic Workflows: A Comprehensive Reality Check. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.2036

@inproceedings{lu2026bitter,
  title = {{The Bitter Lesson of Diffusion Language Models for Agentic Workflows: A Comprehensive Reality Check}},
  author = {Qingyu Lu, Liang Ding, Kanjian Zhang, Jinxia Zhang and Dacheng Tao},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.2036},
  url = {https://doi.org/10.18653/v1/2026.acl-long.2036}
}

References

50 references, 17 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. Gpt-4 technical report. arXiv preprint.
  2. Tomer Amit, Tal Shaharbany, Eliya Nachmani, and Lior Wolf. 2021. Segdiff: Image segmentation with diffusion probabilistic models. arXiv preprint. find
  3. Marianne Arriola, Aaron Gokaslan, Justin T Chiu, Jiaqi Han, Zhihan Yang, Zhixuan Qi, Subham Sekhar Sahoo, and Volodymyr Kuleshov. 2025. Interpolating autoregressive and discrete denoising diffusion language models. In ICLR. find
  4. Yuezhou Hu, Harman Singh, Monishwaran Maheswaran, Haocheng Xi, Coleman Hooper, Jintao Zhang, Aditya Tomar, Michael W. Mahoney, Sewon Min, Mehrdad Farajtabar, Kurt Keutzer, Amir Gholami, and Chenfeng Xu. 2026. Residual context diffusion language models. arXiv preprint. find
  5. Daniel Israel, Guy Van den Broeck, and Aditya Grover. 2025. Accelerating diffusion llms via adaptive parallel decoding. In NeurIPS. find
  6. Albert Q Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, et al. 2024. Mixtral of experts. arXiv preprint. find
  7. Carlos E. Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan. 2023. Swe-bench: Can language models resolve real-world github issues? arXiv preprint.
  8. Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph Gonzalez, Hao Zhang, and Ion Stoica. 2023. Efficient memory management for large language model serving with pagedattention. In SOSP.
  9. Ma Chang, Junlei Zhang, Zhihao Zhu, Cheng Yang, Yujiu Yang, Yaohui Jin, Zhenzhong Lan, Lingpeng Kong, and Junxian He. 2024. Agentboard: An analytical evaluation board of multi-turn llm agents. NeurIPS. find
  10. Jonggeun Lee, Woojung Song, Jongwook Han, Hae-sung Pyun, and Yohan Jo. 2025. Don’t adapt small language models for tools; adapt tool schemas to the models. arXiv preprint. find
  11. Zehui Chen, Kuikun Liu, Qiuchen Wang, Wenwei Zhang, Jiangning Liu, Dahua Lin, Kai Chen, and Feng Zhao. 2024. Agent-FLAN: Designing data and methods of effective agent tuning for large language models. In ACL.
  12. Bill Yuchen Lin, Yicheng Fu, Karina Yang, Faeze Brahman, Shiyu Huang, Chandra Bhagavatula, Prithviraj Ammanabrolu, Yejin Choi, and Xiang Ren. 2023. Swiftsage: A generative agent with fast and slow thinking for complex interactive tasks. NeurIPS.
  13. Maxime Chevalier-Boisvert, Dzmitry Bahdanau, Salem Lahlou, Lucas Willems, Chitwan Saharia, Thien Huu Nguyen, and Yoshua Bengio. 2019. Babyai: A platform to study the sample efficiency of grounded language learning. In ICLR. find
  14. Weiwen Liu, Xu Huang, Xingshan Zeng, Shuai Yu, Dexun Li, Shuai Wang, Weinan Gan, Zhengying Liu, Yuanqing Yu, Zezhong WANG, et al. 2025a. Toolace: Winning the points of llm function calling. In ICLR. find
  15. Liang Ding, Longyue Wang, Xuebo Liu, Derek F Wong, Dacheng Tao, and Zhaopeng Tu. 2021. Understanding and improving lexical choice in non-autoregressive translation. In ICLR. find
  16. Zhaohan Feng, Ruiqi Xue, Lei Yuan, Yang Yu, Ning Ding, Meiqin Liu, Bingzhao Gao, Jian Sun, Xinhu Zheng, and Gang Wang. 2025. Multi-agent embodied ai: Advances and future directions. arXiv preprint. find
  17. Dayuan Fu, Keqing He, Yejie Wang, Wentao Hong, Zhuoma GongQue, Weihao Zeng, Wei Wang, Jingang Wang, Xunliang Cai, and Weiran Xu. 2025a. Agentrefine: Enhancing agent generalization through refinement tuning. In ICLR. find
  18. Dayuan Fu, Jianzhao Huang, Siyuan Lu, Guanting Dong, Yejie Wang, Keqing He, and Weiran Xu. 2025b. PreAct: Prediction enhances agent‘s planning ability. In COLING. find
  19. Yanming Liu, Xinyue Peng, Jiannan Cao, Yuwei Zhang, Xuhong Zhang, Sheng Cheng, Xun Wang, Jianwei Yin, and Tianyu Du. 2025b. Tool-planner: Task planning with clusters across multiple tools. In ICLR. find
  20. Qingyu Lu, Liang Ding, Siyi Cao, Xuebo Liu, Kanjian Zhang, Jinxia Zhang, and Dacheng Tao. 2025. Runaway is ashamed, but helpful: On the early-exit behavior of large language model-based agents in embodied environments. In EMNLP. find
  21. Qingyu Lu, Baopu Qiu, Liang Ding, Kanjian Zhang, Tom Kocmi, and Dacheng Tao. 2024. Error analysis prompting enables human-like translation evaluation in large language models. In ACL.
  22. Elias Lumer, Anmol Gulati, Faheem Nizar, Dzmitry Hedroits, Atharva Mehta, Henry Hwangbo, Vamse Kumar Subbiah, Pradeep Honaganahalli Basavaraju, and James A. Burke. 2025. Tool and agent selection for large language model agents in production: A survey. Preprints. find
  23. Junyu Luo, Weizhi Zhang, Ye Yuan, Yusheng Zhao, Junwei Yang, Yiyang Gu, Bohan Wu, Binqi Chen, Ziyue Qiao, Qingqing Long, et al. 2025. Large language model agent: A survey on methodology, applications and challenges. arXiv preprint.
  24. Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, Shashank Gupta, Bodhisattwa Prasad Majumder, Katherine Hermann, Sean Welleck, Amir Yazdanbakhsh, and Peter Clark. 2023. Self-refine: Iterative refinement with… find
  25. Qingyue Wang, Yanhe Fu, Yanan Cao, Shuai Wang, Zhiliang Tian, and Liang Ding. 2025a. Recursively summarizing enables long-term dialogue memory in large language models. Neurocomputing. find
  26. Ruoyao Wang, Peter Jansen, Marc-Alexandre Côté, and Prithviraj Ammanabrolu. 2022. ScienceWorld: Is your agent smarter than a 5th grader? In EMNLP. find
  27. Xu Wang, Chenkai Xu, Yijie Jin, Jiachun Jin, Hao Zhang, Kai Yu, and Zhijie Deng. 2025b. Diffusion llms can do faster-than-ar inference via discrete diffusion forcing. arXiv preprint. find
  28. KiHyun Nam, Jongmin Choi, Hyeongkeun Lee, Jung-woo Heo, and Joon Son Chung. 2025. Diffusion-link: Diffusion probabilistic model for bridging the audio-text modality gap. arXiv preprint. find
  29. Yu Wang, Ryuichi Takanobu, Zhiqi Liang, Yuzhen Mao, Yuanzhe Hu, Julian McAuley, and Xiaojian Wu. 2025c. Mem-{\\alpha}: Learning memory construction via reinforcement learning. arXiv preprint. find
  30. Shen Nie, Fengqi Zhu, Zebin You, Xiaolu Zhang, Jingyang Ou, Jun Hu, Jun Zhou, Yankai Lin, Ji-Rong Wen, and Chongxuan Li. 2025. Large language diffusion models. NeurIPS.
  31. Chengyue Wu, Hao Zhang, Shuchen Xue, Shizhe Diao, Yonggan Fu, Zhijian Liu, Pavlo Molchanov, Ping Luo, Song Han, and Enze Xie. 2025a. Fast-dllm v2: Efficient block-diffusion llm. arXiv preprint. find
  32. Jiayi Pan, Yichi Zhang, Nicholas Tomlin, Yifei Zhou, Sergey Levine, and Alane Suhr. 2024. Autonomous evaluation and refinement of digital agents. In COLM. find
  33. Chengyue Wu, Hao Zhang, Shuchen Xue, Zhijian Liu, Shizhe Diao, Ligeng Zhu, Ping Luo, Song Han, and Enze Xie. 2025b. Fast-dllm: Training-free acceleration of diffusion llm by enabling kv cache and parallel decoding. arXiv preprint. find
  34. Shishir G. Patil, Huanzhi Mao, Charlie Cheng-Jie Ji, Fanjia Yan, Vishnu Suresh, Ion Stoica, and Joseph E. Gonzalez. 2025. The berkeley function calling leaderboard (bfcl): From tool use to agentic evaluation of large language models. In ICML. find
  35. Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language agents with verbal reinforcement learning. NeurIPS.
  36. Mohit Shridhar, Xingdi Yuan, Marc-Alexandre Cote, Yonatan Bisk, Adam Trischler, and Matthew Hausknecht. 2021. Alfworld: Aligning text and embodied environments for interactive learning. In ICLR.
  37. Yingte Shu, Yuchuan Tian, Chao Xu, Yunhe Wang, and Hanting Chen. 2026. Deferred commitment decoding for diffusion language models. arXiv preprint. find
  38. Yifan Song, Da Yin, Xiang Yue, Jie Huang, Sujian Li, and Bill Yuchen Lin. 2024. Trial and error: Exploration-based trajectory optimization of LLM agents. In ACL.
  39. Zeyi Sun, Ziyu Liu, Yuhang Zang, Yuhang Cao, Xiaoyi Dong, Tong Wu, Dahua Lin, and Jiaqi Wang. 2025. Seagent: Self-evolving computer use agent with autonomous learning from experience. arXiv preprint. find
  40. Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, Jingsen Zhang, Zhiyuan Chen, Jiakai Tang, Xu Chen, Yankai Lin, et al. 2024. A survey on large language model based autonomous agents. Frontiers of Computer Science.
  41. Di Wu, Liang Ding, Fan Lu, and Jian Xie. 2020. Slotre-fine: A fast non-autoregressive model for joint intent detection and slot filling. In EMNLP. find
  42. Yiran Wu, Tianwei Yue, Shaokun Zhang, Chi Wang, and Qingyun Wu. 2024. Stateflow: Enhancing llm task-solving through state-driven workflows. In COLM.
  43. Wujiang Xu, Zujie Liang, Kai Mei, Hang Gao, Juntao Tan, and Yongfeng Zhang. 2025. A-mem: Agentic memory for llm agents. NeurIPS.
  44. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. 2025a. Qwen3 technical report. arXiv preprint.
  45. Ruihan Yang, Prakhar Srivastava, and Stephan Mandt. 2023. Diffusion probabilistic modeling for video generation. Entropy. find
  46. Yicun Yang, Cong Wang, Shaobo Wang, Zichen Wen, Biqing Qi, Hanlin Xu, and Linfeng Zhang. 2025b. Diffusion llm with native variable generation lengths: Let [eos] lead the way. arXiv preprint. find
  47. Shunyu Yao, Noah Shinn, Pedram Razavi, and Karthik Narasimhan. 2024. τ -bench: A benchmark for tool-agent-user interaction in real-world domains. arXiv preprint. find
  48. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. React: Synergizing reasoning and acting in language models. In ICLR.
  49. Jiacheng Ye, Zhihui Xie, Lin Zheng, Jiahui Gao, Zirui Wu, Xin Jiang, Zhenguo Li, and Lingpeng Kong. 2025. Dream 7b: Diffusion large language models. arXiv preprint. find
  50. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, et al. 2023. Judging llm-as-a-judge with mt-bench and chatbot arena. NeurIPS.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/lu-2026-bitter-lesson-diffusion-language-models/