Elamur Research⌕
☰
home / papers / 2026 / ARM2: Adaptive Reasoning Model with Vision Understanding and Executable Code

ARM2: Adaptive Reasoning Model with Vision Understanding and Executable Code

Paper · 2026 · Jian Xie, Zhendong Chu, Aoxiao Zhong et al.

Summary

Reasoning models trained with reinforcement learning tend to "over-think": they produce long chains of thought even for trivial questions, which wastes tokens and can lower accuracy. Earlier fixes, such as fixed token budgets or switching between models, need prior knowledge of task difficulty and do not generalize well. ARM2 is a single 7B model that picks among five reasoning formats for each input. The formats are Direct Answer, Short CoT, Code-Text, Code-Exec (which runs code through a Python interpreter) and Long CoT. ARM2 also accepts images.

Training has two stages. First, a cold-start SFT stage uses 15.1K curated text and vision examples. Second, an RL stage uses GRPO-ALP, a GRPO variant that does two things:

  • it rewards formats that are used less often, which prevents format collapse;
  • it adds a length penalty.

Across six in-domain and six out-of-domain benchmarks, ARM2 roughly matches a GRPO-trained reasoning model while using over 70% fewer tokens on average. This suggests adaptive, length-aware format selection is a practical route to efficient multimodal reasoning.

Key points

  • SFT data (15.1K examples): built from AQuA-Rat (text) and VisualWebInstruct (vision). GPT-4o and Doubao-Pro-1.6-Thinking generated the extra Code-Text, Code-Exec and Long CoT rationales. All code was checked for executability, and rationales leading to wrong answers were removed.
  • RL data (23.2K questions): CommonsenseQA, GSM8K, AIME 1987–2023, Geometry3K, MME RealWorld and MMK12.
  • Length penalty strength: the weight λ was tested from 0.25 to 1, and 0.5 was chosen. Stronger penalties help easy or perception tasks (CSQA, MMEWorld) but hurt multi-step tasks (GSM8K, MMK12).
  • Executable code: adding a real interpreter raised both how often code reasoning was chosen and its accuracy on out-of-domain sets. Majority-vote test-time scaling closed the remaining gaps on GSM8K and GEO3K.
  • Backbone comparison: Mimo-SFT and Mimo-RL backbones gave about 10% higher training accuracy than Qwen. Mimo-RL was unstable on multimodal data, and a short GRPO warm-up added little.

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

🎓The Ohio State UniversityColumbus · United States90 works · 19,112 citations🎓City St George's, University of LondonLondon · United Kingdom3 works · 6 citations

 Fields, topics and concepts

📚 Shelves 5The 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 worksComputer Vision and Pattern Recognition · 3,486 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,685 works

Cite

Cite as Jian Xie, Zhendong Chu, Aoxiao Zhong, Kai Zhang, Mingzhe Han, Xing Fan, Jialie Shen and Qingsong Wen. 2026. ARM2: Adaptive Reasoning Model with Vision Understanding and Executable Code. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1365

@inproceedings{xie2026arm2,
  title = {{ARM2: Adaptive Reasoning Model with Vision Understanding and Executable Code}},
  author = {Jian Xie, Zhendong Chu, Aoxiao Zhong, Kai Zhang, Mingzhe Han, Xing Fan, Jialie Shen and Qingsong Wen},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1365},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1365}
}

References

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

  1. Pranjal Aggarwal and Sean Welleck. 2025. L1: Controlling how long a reasoning model thinks with reinforcement learning. arXiv preprint arXiv:2503.04697.
  2. AIME. Aime problems and solutions. Mohammad Ali Alomrani, Yingxue Zhang, Derek Li, Qianyi Sun, Soumyasundar Pal, Zhanguang Zhang, Yaochen Hu, Rohan Deepak Ajwani, Antonios Valkanas, Raika Karimi, and 1 others. 2025. Reasoning on a budget: A survey of adaptive and controllable test-time compute in ll… — [arXiv:2507.02076](https://arxiv.org/abs/2507.02076) find
  3. Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, and 1 others. 2025. Qwen2. 5-vl technical report. arXiv preprint arXiv:2502.13923. — Qwen2.5-VL Technical Report
  4. Wenhu Chen, Xueguang Ma, Xinyi Wang, and William W Cohen. 2023. Program of thoughts prompting: Disentangling computation from reasoning for numerical reasoning tasks. Transactions on Machine Learning Research. find
  5. Xingyu Chen, Jiahao Xu, Tian Liang, Zhiwei He, Jianhui Pang, Dian Yu, Linfeng Song, Qiuzhi Liu, Mengfei Zhou, Zhuosheng Zhang, and 1 others. 2024. Do not think that much for 2+ 3=? on the overthinking of o1-like llms. arXiv preprint arXiv:2412.21187.
  6. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, and 1 others. 2021. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168.
  7. Alejandro Cuadron, Dacheng Li, Wenjie Ma, Xingyao Wang, Yichuan Wang, Siyuan Zhuang, Shu Liu, Luis Gaspar Schroeder, Tian Xia, Huanzhi Mao, and 1 others. 2025. The danger of overthinking: Examining the reasoning-action dilemma in agentic tasks. arXiv preprint arXiv:2502.08235. find
  8. Chenrui Fan, Ming Li, Lichao Sun, and Tianyi Zhou. 2025. Missing premise exacerbates overthinking: Are reasoning models losing critical thinking skill? In Second Conference on Language Modeling. find
  9. Xingyu Fu, Yushi Hu, Bangzheng Li, Yu Feng, Haoyu Wang, Xudong Lin, Dan Roth, Noah A Smith, Wei-Chiu Ma, and Ranjay Krishna. 2024. Blink: Multimodal large language models can see but not perceive. In European Conference on Computer Vision, pages 148–166. Springer.
  10. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, and 1 others. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948.
  11. Bairu Hou, Yang Zhang, Jiabao Ji, Yujian Liu, Kaizhi Qian, Jacob Andreas, and Shiyu Chang. 2025. Thinkprune: Pruning long chain-of-thought of llms via reinforcement learning. arXiv preprint arXiv:2504.01296.
  12. Edward J Hu, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, and 1 others. 2022. Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations.
  13. Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, Aleksander Madry, Alex Beutel, Alex Carney, and 1 others. 2024. Openai o1 system card. arXiv preprint arXiv:2412.16720.
  14. Yiming Jia, Jiachen Li, Xiang Yue, Bo Li, Ping Nie, Kai Zou, and Wenhu Chen. 2025. Visualwebinstruct: Scaling up multimodal instruction data through web search. arXiv preprint arXiv:2503.10582.
  15. Hunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. 2023. Let’s verify step by step. In The Twelfth International Conference on Learning Representations.
  16. Wang Ling, Dani Yogatama, Chris Dyer, and Phil Blunsom. 2017. Program induction by rationale generation: Learning to solve and explain algebraic word problems. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 158–167. find
  17. Ilya Loshchilov and Frank Hutter. 2017. Sgdr: Stochastic gradient descent with warm restarts. In International Conference on Learning Representations. find
  18. Chenwei Lou, Zewei Sun, Xinnian Liang, Meng Qu, Wei Shen, Wenqi Wang, Yuntao Li, Qingping Yang, and Shuangzhi Wu. 2025. Adacot: Pareto-optimal adaptive chain-of-thought triggering via reinforcement learning. arXiv preprint arXiv:2505.11896. — [arXiv:2505.11896](https://arxiv.org/abs/2505.11896) find
  19. 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. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th Int…
  20. Haotian Luo, Haiying He, Yibo Wang, Jinluan Yang, Rui Liu, Naiqiang Tan, Xiaochun Cao, Dacheng Tao, and Li Shen. 2025a. Ada-r1: Hybrid-cot via bilevel adaptive reasoning optimization. arXiv preprint arXiv:2504.21659. — [arXiv:2504.21659](https://arxiv.org/abs/2504.21659) find
  21. Haotian Luo, Li Shen, Haiying He, Yibo Wang, Shiwei Liu, Wei Li, Naiqiang Tan, Xiaochun Cao, and Dacheng Tao. 2025b. O1-pruner: Length-harmonizing fine-tuning for o1-like reasoning pruning. arXiv preprint arXiv:2501.12570. find
  22. Ahmed Masry, Xuan Long Do, Jia Qing Tan, Shafiq Joty, and Enamul Hoque. 2022. Chartqa: A benchmark for question answering about charts with visual and logical reasoning. In Findings of the Association for Computational Linguistics: ACL 2022, pages 2263– 2279.
  23. Fanqing Meng, Lingxiao Du, Zongkai Liu, Zhixiang Zhou, Quanfeng Lu, Daocheng Fu, Botian Shi, Wenhai Wang, Junjun He, Kaipeng Zhang, and 1 others. 2025. Mm-eureka: Exploring visual aha moment with rule-based large-scale reinforcement learning. arXiv preprint arXiv:2503.07365. find
  24. Todor Mihaylov, Peter Clark, Tushar Khot, and Ashish Sabharwal. 2018. Can a suit of armor conduct electricity? a new dataset for open book question answering. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 2381–2391. find
  25. OpenAI. 2024. Hello GPT-4o. find
  26. David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R Bowman. 2024. Gpqa: A graduate-level google-proof q&a benchmark. In First Conference on Language Modeling.
  27. ByteDance Seed, Jiaze Chen, Tiantian Fan, Xin Liu, Lingjun Liu, Zhiqi Lin, Mingxuan Wang, Chengyi Wang, Xiangpeng Wei, Wenyuan Xu, and 1 others. 2025. Seed1. 5-thinking: Advancing superb reasoning models with reinforcement learning. arXiv preprint arXiv:2504.13914. — Seed1.5-Thinking: Advancing Superb Reasoning Models with Reinforcement Learning
  28. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, and 1 others. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300.
  29. Guangming Sheng, Chi Zhang, Zilingfeng Ye, Xibin Wu, Wang Zhang, Ru Zhang, Yanghua Peng, Haibin Lin, and Chuan Wu. 2024. Hybridflow: A flexible and efficient rlhf framework. arXiv preprint arXiv: 2409.19256.
  30. Yang Sui, Yu-Neng Chuang, Guanchu Wang, Jiamu Zhang, Tianyi Zhang, Jiayi Yuan, Hongyi Liu, Andrew Wen, Shaochen Zhong, Hanjie Chen, and 1 others. 2025. Stop overthinking: A survey on efficient reasoning for large language models. arXiv preprint arXiv:2503.16419.
  31. Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. 2019. Commonsenseqa: A question answering challenge targeting commonsense knowledge. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,…
  32. Xingyao Wang, Yangyi Chen, Lifan Yuan, Yizhe Zhang, Yunzhu Li, Hao Peng, and Heng Ji. 2024. Executable code actions elicit better llm agents. In Forty-first International Conference on Machine Learning.
  33. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc V Le, Ed H Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023. Self-consistency improves chain of thought reasoning in language models. In The Eleventh International Conference on Learning Representations.
  34. Siye Wu, Jian Xie, Jiangjie Chen, Tinghui Zhu, Kai Zhang, and Yanghua Xiao. 2024. How easily do irrelevant inputs skew the responses of large language models? In First Conference on Language Modeling. find
  35. Siye Wu, Jian Xie, Yikai Zhang, Aili Chen, Kai Zhang, Yu Su, and Yanghua Xiao. 2025. Arm: Adaptive reasoning model. arXiv preprint arXiv:2505.20258. — [arXiv:2505.20258](https://arxiv.org/abs/2505.20258) find
  36. LLM-Core-Team Xiaomi. 2025. Mimo-vl technical report. Preprint, arXiv:2506.03569.
  37. Linan Yue, Yichao Du, Yizhi Wang, Weibo Gao, Fangzhou Yao, Li Wang, Ye Liu, Ziyu Xu, Qi Liu, Shimin Di, and 1 others. 2025. Don’t overthink it: A survey of efficient r1-style large reasoning models. arXiv preprint arXiv:2508.02120. find
  38. Xiang Yue, Yuansheng Ni, Kai Zhang, Tianyu Zheng, Ruoqi Liu, Ge Zhang, Samuel Stevens, Dongfu Jiang, Weiming Ren, Yuxuan Sun, and 1 others. 2024. Mmmu: A massive multi-discipline multimodal understanding and reasoning benchmark for expert agi. In Proceedings of the IEEE/CVF Conference on Computer Vi…
  39. Jiajie Zhang, Nianyi Lin, Lei Hou, Ling Feng, and Juanzi Li. 2025. Adaptthink: Reasoning models can learn when to think. arXiv preprint arXiv:2505.13417.
  40. YiFan Zhang, Huanyu Zhang, Haochen Tian, Chaoyou Fu, Shuangqing Zhang, Junfei Wu, Feng Li, Kun Wang, Qingsong Wen, Zhang Zhang, and 1 others. 2024. Mme-realworld: Could your multimodal llm challenge high-resolution real-world scenarios that are difficult for humans? In The Thirteenth International C… find
  41. Jason Zhu and Hongyu Li. 2025. Towards concise and adaptive thinking in large reasoning models: A survey. arXiv preprint arXiv:2507.09662. — [arXiv:2507.09662](https://arxiv.org/abs/2507.09662) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xie-2026-arm2-adaptive-reasoning-model-vision/