Elamur Research⌕
☰
home / papers / 2025 / AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset

AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset

Paper · 2025 · Ivan Moshkov, Darragh Hanley, Ivan Sorokin et al.

Summary

The paper describes the first-place entry in AIMO-2, a Kaggle contest in which systems must answer 50 unpublished national-level math problems under tight time and compute budgets. The authors address weaknesses of open reasoning models in three ways. First, they mine the Art of Problem Solving (AoPS) forums with a pipeline driven by Qwen2.5-32B-Instruct. The pipeline extracts problems, filters them, rewrites proofs as answer-based tasks and decontaminates them. Long chain-of-thought solutions are then distilled from DeepSeek-R1 and QwQ-32B. Second, because those models could not be prompted to call Python, the team bootstraps Tool-Integrated Reasoning (TIR) from LIMO-Qwen-32B. They repeatedly train, generate and keep only samples where code really helps. Third, they train GenSelect, a generative selector that chooses the best of several candidates and outperforms majority voting. The resulting OpenMath-Nemotron models (1.5B–32B, on Qwen2.5-Base) support CoT, TIR and GenSelect modes and set a new open-weight state of the art. Releasing the data, models and code permissively makes the recipe reusable.

Key points

  • The OpenMathReasoning dataset contains 540K unique problems, 3.2M long CoT solutions, 1.7M TIR solutions and 566K GenSelect traces. The problems were reduced from about 620K AoPS discussions through extraction, filtering and decontamination.
  • The winning Kaggle submission used an intermediate 14B model with inference optimizations and solved 34 of 50 private test problems. GenSelect was left out because of the competition's time limits.
  • The authors built a new validation benchmark, Comp-Math-24-25, with 256 problems from AIME 2024/2025 and HMMT 2024/2025.
  • The TIR pipeline includes a mechanism for controlling how many code executions the model may perform in each generation.
  • The code (NeMo-Skills), the models and the data are published on Hugging Face/GitHub under a commercially permissive license.

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 7The field of science this work belongs to (OpenAlex fields).Computer Science · 19,503 worksArtificial Intelligence · 18,805 worksArts and Humanities · 17,959 worksLanguage and Linguistics · 17,789 worksMathematics · 233 worksComputational Theory and Mathematics · 496 worksApplied Mathematics · 73 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,948 works

Cite

Cite as Ivan Moshkov, Darragh Hanley, Ivan Sorokin, Shubham Toshniwal, Christof Henkel, Benedikt Schifferer, Wei Du and Igor Gitman. 2025. AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset. arXiv:2504.16891.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2504.16891 · arXiv 2504.16891

@misc{moshkov2025aimo,
  title = {{AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset}},
  author = {Ivan Moshkov, Darragh Hanley, Ivan Sorokin, Shubham Toshniwal, Christof Henkel, Benedikt Schifferer, Wei Du and Igor Gitman},
  year = {2025},
  doi = {10.48550/arxiv.2504.16891},
  eprint = {2504.16891},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2504.16891}
}

References

43 references, 26 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. [1] Zachary Ankner, Mansheej Paul, Brandon Cui, Jonathan D Chang, and Prithviraj Ammanabrolu. Critique-out-Loud Reward Models. arXiv preprint arXiv:2408.11791, 2024.
  2. [2] bloc97. Ntk-aware scaled rope allows llama models to have extended (8k+) context size without any fine-tuning and minimal perplexity degradation. https://www.reddit.com/r/LocalLLaMA/ comments/14lz7j5/ntkaware_scaled_rope_ allows_llama_models_to_have/, 2023. ↗
  3. [3] Wenhu Chen, Xueguang Ma, Xinyi Wang, and William W. Cohen. Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Tasks. TMLR, 2023. find
  4. [4] Yunfei Cheng, Aonan Zhang, Xuanyu Zhang, Chong Wang, and Yi Wang. Recurrent Drafter for Fast Speculative Decoding in Large Language Models. arXiv preprint arXiv:2403.09919, 2024. find
  5. [5] Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. Training Verifiers to Solve Math Word Problems. arXiv preprint arXiv:2110.14168, 2021.
  6. [6] Long Phan et al. Humanity’s last exam, 2025. — Humanity's Last Exam
  7. [13] Jia Li, Edward Beeching, Lewis Tunstall, Ben Lipkin, Roman Soletskyi, Shengyi Huang, Kashif Rasul, Longhui Yu, Albert Q Jiang, Ziju Shen, et al. NuminaMath: The largest public dataset in AI4Maths with 860k pairs of competition math problems and solutions, 2024. find
  8. [14] Jia LI, Edward Beeching, Lewis Tunstall, Ben Lipkin, Roman Soletskyi, Shengyi Costa Huang, Kashif Rasul, Longhui Yu, Albert Jiang, Ziju Shen, Zihan Qin, Bin Dong, Li Zhou, Yann Fleureau, Guillaume Lample, and Stanislas Polu. NuminaMath. [https://huggingface. co/AI-MO/NuminaMath-1.5](https://git… find
  9. [7] Simon Frieder, Sam Bealing, Arsenii Nikolaiev, Geoff C. Smith, Kevin Buzzard, Timothy Gowers, Peter J. Liu, Po-Shen Loh, Lester Mackey, Leonardo de Moura, Dan Roberts, D. Sculley, Terence Tao, David Balduzzi, Simon Coyle, Alex Gerko, Ryan [15] Xuefeng Li, Haoyang Zou, and Pengfei Liu. Torl: Scal… — ToRL: Scaling Tool-Integrated RL
  10. MARIO: MAth Reasoning 2024. Kaggle. with code Interpreter Output–A Reproducible [8] Silin Gao, Jane Dwivedi-Yu, Ping Yu, Xiaoqing Ellen Tan, Ramakanth Pasunuru, Olga Golovneva, Koustuv Sinha, Asli Celikyilmaz, Antoine Bosselut, and Tianlu Wang. Efficient Tool Use with Chain-of-Abstraction Reasoning.…
  11. [9] Charles Goddard, Shamane Siriwardhana, Malikeh Ehghaghi, Luke Meyers, Vladimir Karpukhin, Brian Benedict, Mark McQuade, and Jacob Solawetz. Arcee’s MergeKit: A Toolkit for Merging Large Language Models, 2024. — Arcee's MergeKit: A Toolkit for Merging Large Language Models
  12. [10] Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning. arXiv preprint arXiv:2501.12948, 2025.
  13. [11] Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. Measuring Mathematical Problem Solving With the MATH Dataset. In NeurIPS Datasets and Benchmarks, 2021.
  14. [12] Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, Aleksander Madry, Alex Beutel, Alex Carney, et al. Openai o1 system card. arXiv preprint arXiv:2412.16720, 2024.
  15. Pipeline. arXiv preprint arXiv:2401.08190, 2024.
  16. [17] Cong Liu, Zhong Wang, ShengYu Shen, Jialiang Peng, Xiaoli Zhang, ZhenDong Du, and YaFang Wang. The chinese dataset distilled from deepseek-r1-671b. https://huggingface.co/datasets/Congliu/ Chinese-DeepSeek-R1-Distill-data-110k, 2025. ↗
  17. [18] Ilya Loshchilov and Frank Hutter. Decoupled Weight Decay Regularization. In ICLR, 2019. find
  18. [19] Dakota Mahan, Duy Van Phung, Rafael Rafailov, and Chase Blagden1 Nathan Lile1 Louis Castricato. Generative Reward Models. arXiv preprint arXiv:2410.12832, 2024.
  19. [20] Sadegh Mahdavi, Muchen Li, Kaiwen Liu, Christos Thrampoulidis, Leonid Sigal, and Renjie Liao. Leveraging Online Olympiad-Level Math Problems for LLMs Training and Contamination-Resistant Evaluation. arXiv preprint arXiv:2501.14275, 2025.
  20. [21] Niklas Muennighoff, Zitong Yang, Weijia Shi, Xiang Lisa Li, Li Fei-Fei, Hannaneh Hajishirzi, Luke Zettlemoyer, Percy Liang, Emmanuel Candès, and Tatsunori Hashimoto. s1: Simple test-time scaling. arXiv preprint arXiv:2501.19393, 2025.
  21. [22] NVIDIA. NeMo-Skills. https://github.com/ NVIDIA/NeMo-Skills. [23] NVIDIA. TensorRT-LLM. https://github. com/NVIDIA/TensorRT-LLM. [24] NVIDIA. TensorRT-LLM Best Practices. https://nvidia.github.io/TensorRT-LLM/ blogs/quantization-in-TRT-LLM.html. [25] OpenR1 Team. OpenR1 Math 220k, February 2025… ↗
  22. [26] Gerald Shen, Zhilin Wang, Olivier Delalleau, Jiaqi Zeng, Yi Dong, Daniel Egert, Shengyang Sun, Jimmy Zhang, Sahil Jain, Ali Taghibakhshi, et al. Nemo-aligner: Scalable toolkit for efficient model alignment. arXiv preprint arXiv:2405.01481, 2024. find
  23. [27] Jianlin Su, Yu Lu, Shengfeng Pan, Ahmed Murtadha, Bo Wen, and Yunfeng Liu. RoFormer: enhanced transformer with rotary position embedding. arXiv preprint arXiv:2104.09864, 2021.
  24. [28] DolphinR1 Team. Dolphin r1. https://huggingface.co/datasets/ cognitivecomputations/dolphinr1, February 2025. Accessed April 2025. ↗
  25. [29] Qwen Team. QwQ-32B: Embracing the Power of Reinforcement Learning, March 2025. find
  26. [30] Shubham Toshniwal, Wei Du, Ivan Moshkov, Branislav Kisacanin, Alexan Ayrapetyan, and Igor Gitman. OpenMathInstruct-2: Accelerating AI for Math with Massive Open-Source Instruction Data. In ICLR, 2025.
  27. [31] Shubham Toshniwal, Ivan Moshkov, Sean Naren-thiran, Daria Gitman, Fei Jia, and Igor Gitman. OpenMathInstruct-1: A 1.8 Million Math Instruction Tuning Dataset. In NeurIPS Datasets and Benchmarks, 2024.
  28. [32] Peifeng Wang, Austin Xu, Yilun Zhou, Caiming Xiong, and Shafiq Joty. Direct Judgement Preference Optimization. arXiv preprint 2409.14664, 2024.
  29. [33] Tianlu Wang, Ilia Kulikov, Olga Golovneva, Ping Yu, Weizhe Yuan, Jane Dwivedi-Yu, Richard Yuanzhe Pang, Maryam Fazel-Zarandi, Jason Weston, and Xian Li. Self-Taught Evaluators. arXiv preprint arXiv:2408.02666, 2024.
  30. [34] Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc V. Le, Ed H. Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. Self-Consistency Improves Chain of Thought Reasoning in Language Models. In ICLR, 2023.
  31. [35] Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. NeurIPS, 2022.
  32. [36] Junde Wu, Jiayuan Zhu, and Yuyuan Liu. Agentic Reasoning: Reasoning LLMs with Tools for the Deep Research. arXiv preprint arXiv:2502.04644, 2025. — [arXiv:2502.04644](https://arxiv.org/abs/2502.04644) find
  33. [37] Tianhao Wu, Weizhe Yuan, Olga Golovneva, Jing Xu, Yuandong Tian, Jiantao Jiao, Jason Weston, and Sainbayar Sukhbaatar. Meta-Rewarding Language Models: Self-Improving Alignment with LLM-as-a-Meta-Judge. arXiv preprint arXiv:2407.19594, 2024.
  34. [38] Wei Xiong, Chengshuai Shi, Jiaming Shen, Aviv Rosenberg, Zhen Qin, Daniele Calandriello, Misha Khalman, Rishabh Joshi, Bilal Piot, Mohammad Saleh, et al. Building math agents with multi-turn iterative preference learning. arXiv preprint arXiv:2409.02392, 2024. — [arXiv:2409.02392](https://arxiv.org/abs/2409.02392) find
  35. [39] An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, Keming Lu, Keqin Bao, Kexin Yang, Le Yu, Mei Li, Mingfeng Xue… find
  36. [40] An Yang, Beichen Zhang, Binyuan Hui, Bofei Gao, Bowen Yu, Chengpeng Li, Dayiheng Liu, Jianhong Tu, Jingren Zhou, Junyang Lin, Keming Lu, Mingfeng Xue, Runji Lin, Tianyu Liu, Xingzhang Ren, and Zhenru Zhang. Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvement, 20…
  37. [41] Shuo Yang, Wei-Lin Chiang, Lianmin Zheng, Joseph E. Gonzalez, and Ion Stoica. Rethinking Benchmark and Contamination for Language Models with Rephrased Samples, 2023.
  38. [42] Yixin Ye, Zhen Huang, Yang Xiao, Ethan Chern, Shijie Xia, and Pengfei Liu. LIMO: Less is More for Reasoning. arXiv preprint arXiv:2502.03387, 2025.
  39. [43] Liang Zeng, Liangjun Zhong, Liang Zhao, Tianwen Wei, Liu Yang, Jujie He, Cheng Cheng, Rui Hu, Yang Liu, Shuicheng Yan, Han Fang, and Yahui Zhou. Skywork-Math: Data Scaling Laws for Mathematical Reasoning in Large Language Models – The Story Goes On, 2024. — Skywork-Math: Data Scaling Laws for Mathematical Reasoning in Large Language Models — The Story Goes On
  40. [44] Bo-Wen Zhang, Yan Yan, Lin Li, and Guang Liu. InfinityMATH: A Scalable Instruction Tuning Dataset in Programmatic Mathematical Reasoning. arXiv preprint arXiv:2408.07089, 2024. — [arXiv:2408.07089](https://arxiv.org/abs/2408.07089) find
  41. [45] Lunjun Zhang, Arian Hosseini, Hritik Bansal, Mehran Kazemi, Aviral Kumar, and Rishabh Agarwal. Generative Verifiers: Reward Modeling as Next-Token Prediction. arXiv preprint arXiv:2408.15240, 2024.
  42. [46] Shaowei Zhang and Deyi Xiong. BackMATH: Towards Backward Reasoning for Solving Math Problems Step by Step. In Proceedings of the 31st International Conference on Computational Linguistics: Industry Track, pages 466–482, 2025. find
  43. [47] Han Zhao, Haotian Wang, Yiping Peng, Sitong Zhao, Xiaoyu Tian, Shuaiting Chen, Yunjie Ji, and Xiangang Li. 1.4 Million Open-Source Distilled Reasoning Dataset to Empower Large Language Model Training. arXiv preprint arXiv:2503.19633, 2025. — [arXiv:2503.19633](https://arxiv.org/abs/2503.19633) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/moshkov-2025-aimo-2-winning-solution-building/