Elamur Research⌕
☰
home / papers / 2026 / Mixture-of-Minds: Multi-Agent Reinforcement Learning for Table Understanding

Mixture-of-Minds: Multi-Agent Reinforcement Learning for Table Understanding

Paper · 2026 · Yuhang Zhou, Mingrui Zhang, Ke Li et al.

Summary

Table question answering requires two things: understanding what a question asks and manipulating tables accurately. Current LLM approaches usually do only one well. Models fine-tuned for reasoning tend to miscalculate and hallucinate. Tool-based pipelines (Python/SQL) compute precisely but are rigid and weak at semantic understanding. Mixture-of-Minds, from Meta AI and Emory University, combines both approaches. It splits each task among three cooperating LLM agents:

  • a planner that drafts the reasoning steps,
  • a coder that writes and executes Python to produce intermediate tables,
  • an answerer that turns this evidence into the final response.

Plans and code have no gold annotations, so the authors generate pseudo-gold intermediate supervision. They run MCTS-style rollouts and keep only paths that reach correct answers. Each agent is then trained in sequence with GRPO, using rewards for plan quality, executable code and answer correctness. On TableBench, Qwen3-32B with this framework reaches 62.13% under parallel test-time scaling, which beats GPT-o3-mini. This shows that structured multi-agent workflows with RL can make smaller open models competitive with proprietary systems.

Key points

  • The workflow uses three agents. Each emits a tagged output: a plan, Python code, or an answer. Code execution yields a sequence of intermediate tables that serve as verifiable evidence.
  • Training is sequential (planning → coding → answering) and uses GRPO on MCTS-derived trajectories. Ablations show the full sequence gives the largest gain. Training the coding agent alone helps more modestly.
  • For Qwen3-32B on TableBench (Fact Checking, Numerical Reasoning, Data Analysis), the average rises from 49.02% with direct single-pass inference to 59.90% with Mixture-of-Minds. It reaches 62.13% with 8× parallel scaling, compared with 59.9% for GPT-o3-mini.
  • Two test-time scaling strategies were tested, both with gains across all evaluated models:
    • Parallel scaling samples the planner 8 times at temperature 1.0 and aggregates results by self-consistency. It averages +3.65%.
    • Sequential scaling regenerates code from error messages until it executes. It averages +1.30%.
    • Direct model-only inference gains little from such scaling.
  • The authors list three limitations: the work covers tables only (no charts or multimodal input), the coder supports Python only (not SQL), and GRPO is sensitive to hyperparameters. Code and data are published at github.com/Tonyzhou98/mixture-of-minds.

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

🎓Emory UniversityAtlanta · United States42 works · 39,899 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 works

Cite

Cite as Yuhang Zhou, Mingrui Zhang, Ke Li, Mingyi Wang, Qiao Liu, Qifei Wang, Jiayi Liu, Fei Liu, Serena Li, Weiwei LI, Mingze Gao, Abhishek Kumar, Xiangjun Fan, Zhuokai Zhao and Lizhu Zhang. 2026. Mixture-of-Minds: Multi-Agent Reinforcement Learning for Table Understanding. 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.112

@inproceedings{zhou2026mixture,
  title = {{Mixture-of-Minds: Multi-Agent Reinforcement Learning for Table Understanding}},
  author = {Yuhang Zhou, Mingrui Zhang, Ke Li, Mingyi Wang, Qiao Liu, Qifei Wang, Jiayi Liu, Fei Liu, Serena Li, Weiwei LI, Mingze Gao, Abhishek Kumar, Xiangjun Fan, Zhuokai Zhao and Lizhu Zhang},
  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.112},
  url = {https://doi.org/10.18653/v1/2026.acl-long.112}
}

References

50 references, 24 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Akhiad Bercovich, Itay Levy, Izik Golan, Mohammad Dabbah, Ran El-Yaniv, Omri Puny, Ido Galil, Zach Moshe, Tomer Ronen, Najeeb Nabwani, Ido Shahaf, Oren Tropp, Ehud Karpas, Ran Zilberstein, Jiaqi Zeng, Soumye Singhal, Alexander Bukharin, Yian Zhang, Tugrul Konuk, Gerald Shen, Ameya Sunil Mahabaleshwa… — Llama-Nemotron: Efficient Reasoning Models
  2. Leonardo Bertolazzi, Philipp Mondorf, Barbara Plank, and Raffaella Bernardi. 2025. The validation gap: A mechanistic analysis of how language models compute arithmetic but fail to validate it. arXiv preprint arXiv:2502.11771.
  3. Cameron B Browne, Edward Powley, Daniel Whitehouse, Simon M Lucas, Peter I Cowling, Philipp Rohlfshagen, Stephen Tavener, Diego Perez, Spyridon Samothrakis, and Simon Colton. 2012. A survey of monte carlo tree search methods. IEEE Transactions on Computational Intelligence and AI in games, 4(1):1–43… find
  4. Lei Chen, Xuanle Zhao, Zhixiong Zeng, Jing Huang, Liming Zheng, Yufeng Zhong, and Lin Ma. 2025. Breaking the sft plateau: Multimodal structured reinforcement learning for chart-to-code generation. arXiv preprint arXiv:2508.13587. — [arXiv:2508.13587](https://arxiv.org/abs/2508.13587) find
  5. Linguistics (Volume 1: Long Papers), pages 1821– 1831. find
  6. Wenhu Chen, Hongmin Wang, Jianshu Chen, Yunkai Zhang, Hong Wang, Shiyang Li, Xiyou Zhou, and William Yang Wang. 2019. Tabfact: A large-scale dataset for table-based fact verification. arXiv preprint arXiv:1909.02164.
  7. Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, Aleksander Madry, Alex Beutel, Alex Carney, et al. 2024. Openai o1 system card. arXiv preprint arXiv:2412.16720.
  8. Zhiyu Chen, Wenhu Chen, Charese Smiley, Sameena Shah, Iana Borova, Dylan Langdon, Reema Moussa, Matt Beane, Ting-Hao Huang, Bryan Routledge, and William Yang Wang. 2022. Finqa: A dataset of numerical reasoning over financial data.
  9. Bowen Jin, Hansi Zeng, Zhenrui Yue, Jinsung Yoon, Sercan Arik, Dong Wang, Hamed Zamani, and Jiawei Han. 2025. Search-r1: Training llms to reason and leverage search engines with reinforcement learning. arXiv preprint arXiv:2503.09516.
  10. Zhiyu Chen, Wenhu Chen, Charese Smiley, Sameena Shah, Iana Borova, Dylan Langdon, Reema Moussa, Matt Beane, Ting-Hao Huang, Bryan Routledge, et al. 2021. Finqa: A dataset of numerical reasoning over financial data. arXiv preprint arXiv:2109.00122.
  11. Nengzheng Jin, Joanna Siebert, Dongfang Li, and Qingcai Chen. 2022. A survey on table question answering: recent advances. In China Conference on Knowledge Graph and Semantic Computing, pages 174– 186. Springer. find
  12. Mingyue Cheng, Qingyang Mao, Qi Liu, Yitong Zhou, Yupeng Li, Jiahao Wang, Jiaying Lin, Jiawei Cao, and Enhong Chen. 2025. A survey on table mining with large language models: Challenges, advancements and prospects. Authorea Preprints. find
  13. Adam Tauman Kalai, Ofir Nachum, Santosh S Vempala, and Edwin Zhang. 2025. Why language models hallucinate. arXiv preprint arXiv:2509.04664. — [arXiv:2509.04664](https://arxiv.org/abs/2509.04664) find
  14. Xi Fang, Weijie Xu, Fiona Anting Tan, Jiani Zhang, Ziqing Hu, Yanjun Qi, Scott Nickleach, Diego Socol-insky, Srinivasan Sengamedu, and Christos Faloutsos. 2024. Large language models (llms) on tabular data: Prediction, generation, and understanding–a survey. arXiv preprint arXiv:2402.17944. — [arXiv:2402.17944](https://arxiv.org/abs/2402.17944) find
  15. Jiazhan Feng, Shijue Huang, Xingwei Qu, Ge Zhang, Yujia Qin, Baoquan Zhong, Chengquan Jiang, Jinxin Chi, and Wanjun Zhong. 2025. Retool: Reinforcement learning for strategic tool use in llms. arXiv preprint arXiv:2504.11536.
  16. Yichao Fu, Xuewei Wang, Yuandong Tian, and Jiawei Zhao. 2025. Deep think with confidence. arXiv preprint arXiv:2508.15260.
  17. Yannis Katsis, Saneem Chemmengath, Vishwajeet Kumar, Samarth Bharadwaj, Mustafa Canim, Michael Glass, Alfio Gliozzo, Feifei Pan, Jaydeep Sen, Karthik Sankaranarayanan, et al. 2021. Ait-qa: Question answering dataset over complex tables in the airline industry. arXiv preprint arXiv:2106.12944. find
  18. Fangyu Lei, Jinxiang Meng, Yiming Huang, Tinghong Chen, Yun Zhang, Shizhu He, Jun Zhao, and Kang Liu. 2025. Reasoning-table: Exploring reinforcement learning for table reasoning. arXiv preprint arXiv:2506.01710. — [arXiv:2506.01710](https://arxiv.org/abs/2506.01710) find
  19. Jia Li, Yunfei Zhao, Yongmin Li, Ge Li, and Zhi Jin. 2024. Acecoder: An effective prompting technique specialized in code generation. ACM Transactions on Software Engineering and Methodology, 33(8):1–26. find
  20. Dawei Gao, Haibin Wang, Yaliang Li, Xiuyu Sun, Yichen Qian, Bolin Ding, and Jingren Zhou. 2023. Text-to-sql empowered by large language models: A benchmark evaluation. arXiv preprint arXiv:2308.15363.
  21. Shu Liu, Sumanth Hegde, Shiyi Cao, Alan Zhu, Dacheng Li, Tyler Griggs, Eric Tang, Akshay Malik, Kourosh Hakhamaneshi, Richard Liaw, Philipp Moritz, Matei Zaharia, Joseph E. Gonzalez, and Ion Stoica. 2025. Skyrl-sql: Matching gpt-4o and o4-mini on text2sql with multi-turn rl. find
  22. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783.
  23. Keer Lu, Chong Chen, Bin Cui, Huang Leng, and Wentao Zhang. 2025. Pilotrl: Training language model agents via global planning-guided progressive reinforcement learning. arXiv preprint arXiv:2508.00344. — [arXiv:2508.00344](https://arxiv.org/abs/2508.00344) find
  24. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948.
  25. Thomas Mueller, Francesco Piccinno, Massimo Nicosia, Peter Shaw, and Yasemin Altun. 2019. Answering conversational questions on structured data without logical forms. arXiv preprint arXiv:1908.11787. find
  26. Mohit Iyyer, Wen-tau Yih, and Ming-Wei Chang. 2017. Search-based neural structured learning for sequential question answering. In Proceedings of the 55th Annual Meeting of the Association for Computational Niklas Muennighoff, Zitong Yang, Weijia Shi, Xiang Lisa Li, Li Fei-Fei, Hannaneh Hajishirzi, L…
  27. OpenAI. 2023. Gpt-4 technical report.
  28. Hadas Orgad, Michael Toker, Zorik Gekhman, Roi Reichart, Idan Szpektor, Hadas Kotek, and Yonatan Belinkov. 2024. Llms know more than they show: On the intrinsic representation of llm hallucinations. arXiv preprint arXiv:2410.02707.
  29. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics, pages 311–318.
  30. Panupong Pasupat and Percy Liang. 2015. Compositional semantic parsing on semi-structured tables. arXiv preprint arXiv:1508.00305.
  31. Amit Rath. 2025. Structured prompting and feedback-guided reasoning with llms for data interpretation. arXiv preprint arXiv:2505.01636. — [arXiv:2505.01636](https://arxiv.org/abs/2505.01636) find
  32. Timo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. 2023. Toolformer: Language models can teach themselves to use tools. Advances in Neural Information Processing Systems, 36:68539–68551.
  33. John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. 2017. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347. find
  34. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Y Wu, et al. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300.
  35. Liang Shi, Zhengju Tang, Nan Zhang, Xiaotong Zhang, and Zhi Yang. 2024. A survey on employing large language models for text-to-sql tasks. ACM Computing Surveys. find
  36. Yuan Sui, Mengyu Zhou, Mingjie Zhou, Shi Han, and Dongmei Zhang. 2024. Table meets llm: Can large language models understand structured table data? a benchmark and empirical study. In Proceedings of the 17th ACM International Conference on Web Search and Data Mining, pages 645–654.
  37. Ruoxi Sun, Sercan Ö Arik, Alex Muzio, Lesly Miculicich, Satya Gundabathula, Pengcheng Yin, Hanjun Dai, Hootan Nakhost, Rajarishi Sinha, Zifeng Wang, et al. 2023. Sql-palm: Improved large language model adaptation for text-to-sql (extended). arXiv preprint arXiv:2306.00739. — [arXiv:2306.00739](https://arxiv.org/abs/2306.00739) find
  38. Alon Talmor, Ori Yoran, Amnon Catav, Dan Lahav, Yizhong Wang, Akari Asai, Gabriel Ilharco, Hannaneh Hajishirzi, and Jonathan Berant. 2021. Multimodalqa: Complex question answering over text, tables and images. arXiv preprint arXiv:2104.06039. — [arXiv:2104.06039](https://arxiv.org/abs/2104.06039) find
  39. Zirui Tang, Weizheng Wang, Zihang Zhou, Yang Jiao, Bangrui Xu, Boyu Niu, Xuanhe Zhou, Guoliang Li, Yeye He, Wei Zhou, et al. 2025. Llm/agent-as-data-analyst: A survey. arXiv preprint arXiv:2509.23988. — [arXiv:2509.23988](https://arxiv.org/abs/2509.23988) find
  40. Gemma Team, Aishwarya Kamath, Johan Ferret, Shreya Pathak, Nino Vieillard, Ramona Merhej, Sarah Perrin, Tatiana Matejovicova, Alexandre Ramé, Morgane Rivière, et al. 2025. Gemma 3 technical report. arXiv preprint arXiv:2503.19786. find
  41. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2022. Self-consistency improves chain of thought reasoning in language models. arXiv preprint arXiv:2203.11171.
  42. Zilong Wang, Hao Zhang, Chun-Liang Li, Julian Martin Eisenschlos, Vincent Perot, Zifeng Wang, Lesly Miculicich, Yasuhisa Fujii, Jingbo Shang, Chen-Yu Lee, et al. 2024. Chain-of-table: Evolving tables in the reasoning chain for table understanding. arXiv preprint arXiv:2401.04398.
  43. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in Neural Information Processing Systems, 35:24824–24837.
  44. Mingyuan Wu, Jingcheng Yang, Jize Jiang, Meitang Li, Kaizhuo Yan, Hanchao Yu, Minjia Zhang, Chengxiang Zhai, and Klara Nahrstedt. 2025a. Vtool-r1: Vlms learn to think with images via reinforcement learning on multimodal tool use. arXiv preprint arXiv:2505.19255. — [arXiv:2505.19255](https://arxiv.org/abs/2505.19255) find
  45. Xianjie Wu, Jian Yang, Linzheng Chai, Ge Zhang, Jiaheng Liu, Xeron Du, Di Liang, Daixin Shu, Xianfu Cheng, Tianzhen Sun, et al. 2025b. Tablebench: A comprehensive and complex benchmark for table question answering. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 25… find
  46. Zhenhe Wu, Jian Yang, Jiaheng Liu, Xianjie Wu, Changzai Pan, Jie Zhang, Yu Zhao, Shuangyong Song, Yongxiang Li, and Zhoujun Li. 2025c. Tabler1: Region-based reinforcement learning for table understanding. arXiv preprint arXiv:2505.12415. xAI. 2025. Grok-4. Accessed: 2025-07-10. — [arXiv:2505.12415](https://arxiv.org/abs/2505.12415) find
  47. 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 arXiv:2505.09388.
  48. Chenghao Yang and Ari Holtzman. 2025. How alignment shrinks the generative horizon. arXiv preprint arXiv:2506.17871. — [arXiv:2506.17871](https://arxiv.org/abs/2506.17871) find
  49. Zheyuan Yang, Lyuhao Chen, Arman Cohan, and Yilun Zhao. 2025b. Table-r1: Inference-time scaling for table reasoning. arXiv preprint arXiv:2505.23621. — [arXiv:2505.23621](https://arxiv.org/abs/2505.23621) find
  50. Yunhu Ye, Binyuan Hui, Min Yang, Binhua Li, Fei Huang, and Yongbin Li. 2023. Large language models are versatile decomposers: Decomposing evidence and questions for table-based reasoning. In Proceedings of the 46th international ACM SIGIR conference on research and development in information retriev… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhou-2026-mixture-minds-multi-agent-reinforcement/