Elamur Research⌕
☰
home / papers / 2025 / VisuoThink: Empowering LVLM Reasoning with Multimodal Tree Search

VisuoThink: Empowering LVLM Reasoning with Multimodal Tree Search

Paper · 2025 · Yikun Wang, Siyin Wang, Qinyuan Cheng et al.

Summary

Large vision-language models (LVLMs) still struggle with geometry problems and spatial navigation. People handle such tasks by sketching, checking intermediate pictures and trying alternative routes. Earlier multimodal "slow thinking" methods either reason purely in text over a fixed input image or add visual help only once or in a crude form, such as emojis.

VisuoThink is a training-free inference framework. The model alternates textual reasoning with tool calls, for example Python code that draws auxiliary lines, and inspects the updated image before continuing. On top of this loop, a look-ahead tree search does three things at each step:

  • it samples several candidate next steps;
  • it extends each candidate with a simulated rollout to a final answer;
  • it lets the model choose the best branch by self-voting.

The authors tested the framework on GPT-4o, Claude-3.5-Sonnet and Qwen2-72B-VL, and it achieves state-of-the-art results on geometry and spatial reasoning tasks, including 48.5% accuracy@1 on Geomeverse. The work matters because it shows that test-time scaling can extend to visual reasoning without any fine-tuning.

Key points

  • Search procedure: each step runs a Thought–Action–Observation cycle inside a three-stage search: vision-text interleaved expansion of k candidates, rollout simulation, and selection by self-voting.
  • Geomeverse results: accuracy@1 reaches up to 48.5%, an improvement of up to 21.8% over the strongest prior baseline.
  • Scaling tool calls: allowing more reasoning steps brought large gains up to 20 steps (+54.1% for GPT-4o, +48.4% for Claude-3.5-Sonnet). Going from 20 to 40 steps added only +6.5% and +2.1%.
  • Scaling tree width: adding more candidate child nodes produces an inverted-U performance curve, because the model's evaluation of nodes is imperfect.
  • Supervision strength: rollouts help more in spatial reasoning than in geometry. Spatial rollouts return rich state information, such as the agent and goal positions, while geometry rollouts return only a final number. Limitations include high compute cost and reliance on tools. Code is available at github.com/ekonwang/VisuoThink.

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

 Fields, topics and concepts

📚 Shelves 3The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 2Research subjects and languages the work is about.Reasoning · 1,426 worksVision & language · 657 works

Cite

Cite as Yikun Wang, Siyin Wang, Qinyuan Cheng, Zhaoye Fei, Liang Ding, Qipeng Guo, Dacheng Tao and Xipeng Qiu. 2025. VisuoThink: Empowering LVLM Reasoning with Multimodal Tree Search. Proceedings of the 63rd 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/v1/2025.acl-long.1053

@inproceedings{wang2025visuothink,
  title = {{VisuoThink: Empowering LVLM Reasoning with Multimodal Tree Search}},
  author = {Yikun Wang, Siyin Wang, Qinyuan Cheng, Zhaoye Fei, Liang Ding, Qipeng Guo, Dacheng Tao and Xipeng Qiu},
  booktitle = {Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2025},
  doi = {10.18653/v1/2025.acl-long.1053},
  url = {https://doi.org/10.18653/v1/2025.acl-long.1053}
}

References

41 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Afra Amini, Tim Vieira, and Ryan Cotterell. 2024. Variational best-of-n alignment. ArXiv, abs/2407.06057. — [arXiv:2407.06057](https://arxiv.org/abs/2407.06057) find
  2. Jinhe Bi, Yifan Wang, Danqi Yan, Xun Xiao, Artur Hecker, Volker Tresp, and Yunpu Ma. 2025a. Prism: Self-pruning intrinsic selection method for training-free multimodal data selection. Preprint, arXiv:2502.12119. — [arXiv:2502.12119](https://arxiv.org/abs/2502.12119) find
  3. Jinhe Bi, Yujun Wang, Haokun Chen, Xun Xiao, Artur Hecker, Volker Tresp, and Yunpu Ma. 2025b. Llava steering: Visual instruction tuning with 500x fewer parameters through modality linear representation-steering. Preprint, arXiv:2412.12359.
  4. Guoxin Chen, Minpeng Liao, Chengxi Li, and Kai Fan. 2024. Alphamath almost zero: process supervision without process. ArXiv, abs/2405.03553.
  5. Anoop Cherian, Kuan-Chuan Peng, Suhas Lohit, Joanna Matthiesen, Kevin Smith, and Joshua B Tenenbaum. 2024. Evaluating large vision-and-language models on children’s mathematical olympiads. arXiv preprint arXiv:2406.15736. — [arXiv:2406.15736](https://arxiv.org/abs/2406.15736) find
  6. DeepSeek-AI. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. Preprint, arXiv:2501.12948.
  7. Yifan Du, Zikang Liu, Yifan Li, Wayne Xin Zhao, Yuqi Huo, Bingning Wang, Weipeng Chen, Zheng Liu, Zhongyuan Wang, and Jiahui Wen. 2025. Virgo: A preliminary exploration on reproducing o1-like mllm. find
  8. Xidong Feng, Ziyu Wan, Muning Wen, Ying Wen, Weinan Zhang, and Jun Wang. 2023. Alphazero-like tree-search can guide large language model decoding and training. ArXiv, abs/2309.17179. find
  9. Timin Gao, Peixian Chen, Mengdan Zhang, Chaoyou Fu, Yunhang Shen, Yan Zhang, Shengchuan Zhang, Xiawu Zheng, Xing Sun, Liujuan Cao, and Rongrong Ji. 2024. Cantor: Inspiring multimodal chain-of-thought of mllm. ArXiv, abs/2404.16033. — [arXiv:2404.16033](https://arxiv.org/abs/2404.16033) find
  10. Lin Gui, Cristina Garbacea, and Victor Veitch. 2024. Bonbon alignment for large language models and the sweetness of best-of-n sampling. ArXiv, abs/2406.00832. — [arXiv:2406.00832](https://arxiv.org/abs/2406.00832) find
  11. Yushi Hu, Weijia Shi, Xingyu Fu, Dan Roth, Mari Ostendorf, Luke S. Zettlemoyer, Noah A. Smith, and Ranjay Krishna. 2024. Visual sketchpad: Sketching as a visual chain of thought for multimodal language models. ArXiv, abs/2406.09403. find
  12. Daniel Kahneman. 2011. Thinking, fast and slow. Farrar, Straus and Giroux. find
  13. Mehran Kazemi, Hamidreza Alvari, Ankit Anand, Jialin Wu, Xi Chen, and Radu Soricut. 2023. Geomverse: A systematic evaluation of large models for geometric reasoning. Preprint, arXiv:2312.12241. — [arXiv:2312.12241](https://arxiv.org/abs/2312.12241) find
  14. Xuanyu Lei, Zonghan Yang, Xinrui Chen, Peng Li, and Yang Liu. 2024. Scaffolding coordinates to promote vision-language coordination in large multi-modal models. In International Conference on Computational Linguistics.
  15. Chengzu Li, Wenshan Wu, Huanyu Zhang, Yan Xia, Shaoguang Mao, Li Dong, Ivan Vuli’c, and Furu Wei. 2025. Imagine while reasoning in space: Multimodal visualization-of-thought. find
  16. Jiacheng Liu, Andrew Cohen, Ramakanth Pasunuru, Yejin Choi, Hannaneh Hajishirzi, and Asli Celikyilmaz. 2023. Don’t throw away your value model! generating more preferable text with value-guided monte-carlo tree search decoding.
  17. Pan Lu, Ran Gong, Shibiao Jiang, Liang Qiu, Siyuan Huang, Xiaodan Liang, and Song-Chun Zhu. 2021. Inter-gps: Interpretable geometry problem solving with formal language and symbolic reasoning. Preprint, arXiv:2105.04165.
  18. Chancharik Mitra, Brandon Huang, Trevor Darrell, and Roei Herzig. 2023. Compositional chain-of-thought prompting for large multimodal models. 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 14420–14431. find
  19. Debjyoti Mondal, Suraj Modi, Subhadarshi Panda, Rituraj Singh, and Godawari Sudhakar Rao. 2024. Kam-cot: Knowledge augmented multimodal chain-of-thoughts reasoning. In AAAI Conference on Artificial Intelligence. find
  20. OpenAI. 2024a. Gpt-4o system card. arXiv:2410.21276.
  21. Preprint, OpenAI. 2024b. Learning to reason with llms. find
  22. Runqi Qiao, Qiuna Tan, Guanting Dong, Minhui Wu, Chong Sun, Xiaoshuai Song, Zhuoma GongQue, Shanglin Lei, Zhe Wei, Miaoxuan Zhang, and 1 others. 2024. We-math: Does your large multimodal model achieve human-like mathematical reasoning? arXiv preprint arXiv:2407.01284. find
  23. Santhosh Kumar Ramakrishnan, Erik Wijmans, Philipp Kraehenbuehl, and Vladlen Koltun. 2024. Does spatial cognition emerge in frontier models? Preprint, arXiv:2410.06468. — [arXiv:2410.06468](https://arxiv.org/abs/2410.06468) find
  24. Noah Shinn, Federico Cassano, Edward Berman, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language agents with verbal reinforcement learning. Preprint, arXiv:2303.11366.
  25. Gemini Team. 2024. Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context. Preprint, arXiv:2403.05530.
  26. Omkar Thawakar, Dinura Dissanayake, Ketan More, Ritesh Thawkar, Ahmed Heakl, Noor Ahsan, Yuhao Li, Mohammed Zumri, Jean Lahoud, Rao Muhammad Anwer, Hisham Cholakkal, Ivan Laptev, Mubarak Shah, Fahad Shahbaz Khan, and Salman H. Khan. 2025. Llamav-o1: Rethinking step-by-step visual reasoning in llms. find
  27. Wenshan Wu, Shaoguang Mao, Yadong Zhang, Yan Xia, Li Dong, Lei Cui, and Furu Wei. 2024. Mind’s eye of llms: Visualization-of-thought elicits spatial reasoning in large language models. find
  28. Kun Xiang, Zhili Liu, Zihao Jiang, Yunshuang Nie, Runhui Huang, Haoxiang Fan, Hanhui Li, Weiran Huang, Yihan Zeng, Jianhua Han, Lanqing Hong, Hang Xu, and Xiaodan Liang. 2024. Atomthink: A slow thinking framework for multimodal mathematical reasoning. Preprint, arXiv:2411.11930. — [arXiv:2411.11930](https://arxiv.org/abs/2411.11930) find
  29. Yuxi Xie, Kenji Kawaguchi, Yiran Zhao, Xu Zhao, MingSung Kan, Junxian He, and Qizhe Xie. 2023. Self-evaluation guided beam search for reasoning. In Neural Information Processing Systems.
  30. Guowei Xu, Peng Jin, Hao Li, Yibing Song, Lichao Sun, and Li Yuan. 2024. Llava-cot: Let vision language models reason step-by-step. ArXiv, abs/2411.10440. find
  31. Zhengyuan Yang, Linjie Li, Jianfeng Wang, Kevin Lin, Ehsan Azarnasab, Faisal Ahmed, Zicheng Liu, Ce Liu, Michael Zeng, and Lijuan Wang. 2023. Mm-react: Prompting chatgpt for multimodal reasoning and action. ArXiv, abs/2303.11381. find
  32. Huanjin Yao, Jiaxing Huang, Wenhao Wu, Jingyi Zhang, Yibo Wang, Shunyu Liu, Yingjie Wang, Yuxin Song, Haocheng Feng, Li Shen, and Dacheng Tao. 2024. Mulberry: Empowering mllm with o1-like reasoning and reflection via collective monte carlo tree search. ArXiv, abs/2412.18319. — [arXiv:2412.18319](https://arxiv.org/abs/2412.18319) find
  33. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. React: Synergizing reasoning and acting in language models. Preprint, arXiv:2210.03629.
  34. Fei Yu, Anningzhe Gao, and Benyou Wang. 2023. Ovm, outcome-supervised value models for planning in mathematical reasoning. In NAACL-HLT. find
  35. Charlie Snell, Jaehoon Lee, Kelvin Xu, and Aviral Kumar. 2024. Scaling llm test-time compute optimally can be more effective than scaling model parameters. ArXiv, abs/2408.03314.
  36. Zhiyuan Zeng, Qinyuan Cheng, Zhangyue Yin, Bo Wang, Shimin Li, Yunhua Zhou, Qipeng Guo, Xuanjing Huang, and Xipeng Qiu. 2024. Scaling of search and learning: A roadmap to reproduce o1 from reinforcement learning perspective. Preprint, arXiv:2412.14135. find
  37. Hanshi Sun, Momin Haider, Ruiqi Zhang, Huitao Yang, Jiahao Qiu, Ming Yin, Mengdi Wang, Peter Bartlett, and Andrea Zanette. 2024. Fast best-of-n decoding via speculative rejection. ArXiv, abs/2410.20290. find
  38. Zhuosheng Zhang, Aston Zhang, Mu Li, Hai Zhao, George Karypis, and Alexander J. Smola. 2023. Multimodal chain-of-thought reasoning in language models.
  39. Trans. Mach. Learn. Res., 2024. find
  40. Wenliang Zhao, Yongming Rao, Zuyan Liu, Benlin Liu, Jie Zhou, and Jiwen Lu. 2023. Unleashing text-to-image diffusion models for visual perception. Preprint, arXiv:2303.02153. find
  41. Qiji Zhou, Ruochen Zhou, Zike Hu, Panzhong Lu, Siyang Gao, and Yue Zhang. 2024. Image-of-thought prompting for visual reasoning refinement in multimodal large language models. ArXiv, abs/2405.13872. — [arXiv:2405.13872](https://arxiv.org/abs/2405.13872) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2025-visuothink-empowering-lvlm-reasoning-multimodal/