Elamur Research⌕
☰
home / papers / 2026 / PaCoRe: Learning to Scale Test-Time Compute with Parallel Coordinated Reasoning

PaCoRe: Learning to Scale Test-Time Compute with Parallel Coordinated Reasoning

Paper · 2026 · Jingcheng Hu, Yinmin Zhang, Shijie Shang et al.

Summary

Language models that reason step by step in a single chain hit a hard limit: once the context window is full, reasoning must stop. This caps how much test-time compute (TTC) can be spent on one problem. The authors from Tsinghua University and StepFun propose PaCoRe (Parallel Coordinated Reasoning), which shifts the main source of compute from sequential depth to coordinated breadth. PaCoRe runs in rounds. In each round, many reasoning trajectories run in parallel. Each trajectory is then cut down to its final conclusion, and these short messages, together with the problem, become the input for the next round. The model is trained with large-scale, outcome-based reinforcement learning to synthesize these messages. Without such training, models often fall into "Reasoning Solipsism": they ignore the messages from other branches and solve the problem from scratch. With this setup, effective TTC reaches millions of tokens while each input still fits in the context window. The gains appear across several domains and are largest in math. The released resources make large-scale parallel reasoning, used by closed frontier systems, open to study.

Key points

  • PaCoRe-8B scores 94.5% on HMMT 2025, higher than GPT-5's 93.2%. This result uses about two million tokens of effective TTC per problem.
  • Accuracy on HMMT 2025 increases with the inference configuration: 90.3% with [8], 92.9% with [16], 93.9% with [32], and 94.5% with the multi-round [32, 4] setting.
  • Message compaction removes the intermediate chain-of-thought and keeps only each trajectory's conclusion. The same model weights are used in every round, and the last round produces a single answer.
  • On LiveCodeBench, an RLVR-8B baseline does not benefit from additional parallel compute, whereas PaCoRe-8B does.
  • A Qwen3-1.7B-PaCoRe variant reaches 35.2% on HMMT 2025 with [8], compared with 32.1% reported for AggLM-1.7B. It rises to 41.9% with [32, 4].
  • The authors have publicly released the trained model weights, the training dataset, and their complete pipeline for running inference.
  • The approach has so far been tested only on text: tool use and multimodal input are not yet supported.

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

🎓Tsinghua UniversityBeijing · China333 works · 87,019 citations🎓Peking UniversityBeijing · China244 works · 39,359 citations

 Fields, topics and concepts

📚 Shelves 5The 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 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,426 works

Cite

Cite as Jingcheng Hu, Yinmin Zhang, Shijie Shang, Xiaobo Yang, Yue Peng, Zhewei Huang, Hebin Zhou, Xin Wu, Jie Cheng, Fanqi Wan, Xiangwen Kong, Chengyuan Yao, Kaiwen Yan, Ailin Huang, Hongyu Zhou, Qi Han, Zheng Ge, Xiangyu Zhang and Heung-Yeung Shum. 2026. PaCoRe: Learning to Scale Test-Time Compute with Parallel Coordinated Reasoning. 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.1253

@inproceedings{hu2026pacore,
  title = {{PaCoRe: Learning to Scale Test-Time Compute with Parallel Coordinated Reasoning}},
  author = {Jingcheng Hu, Yinmin Zhang, Shijie Shang, Xiaobo Yang, Yue Peng, Zhewei Huang, Hebin Zhou, Xin Wu, Jie Cheng, Fanqi Wan, Xiangwen Kong, Chengyuan Yao, Kaiwen Yan, Ailin Huang, Hongyu Zhou, Qi Han, Zheng Ge, Xiangyu Zhang and Heung-Yeung Shum},
  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.1253},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1253}
}

References

65 references, 20 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Aime problems and solutions. https://artofproblemsolving.com/wiki/index. php/AIME_Problems_and_Solutions. Accessed: 2025-12-07. ↗
  2. Berkeley math tournament (bmt). https://berkeley.mt/. Accessed: 2025-12-07. ↗
  3. Brown university math olympiad (brumo). https://www.brumo.org/. Accessed: 2025-12-07. ↗
  4. Caltech harvey mudd math competition (chmmc). http://chmmc.caltech.edu/problems. html. Accessed: 2025-12-07. ↗
  5. Carnegie mellon informatics and mathematics competition (cmimc). https://cmimc.math.cmu.edu/. Accessed: 2025-12-07. ↗
  6. Duke math meet (dmm). https://dukemathmeet.org/. Accessed: 2025-12-07. ↗
  7. Harvard–mit mathematics tournament (hmmt). https://www.hmmt.org/. Accessed: 2025-12-07. ↗
  8. Math prize for girls. https://mathprize.atfoundation.org/. Accessed: 2025-12-07. ↗
  9. Minnesota youth math outreach (mnymo). https://www.mnyouthmathoutreach.org/. Accessed: 2025-12-07. ↗
  10. Princeton university mathematics competition (pumac). https://jason-shi-f9dm.squarespace.com/. Accessed: 2025-12-07. ↗
  11. Stanford math tournament (smt). https://www.stanfordmathtournament.org/. Accessed: 2025-12-07. ↗
  12. Milad Aghajohari, Kamran Chitsaz, Amirhossein Kazemnejad, Sarath Chandar, Alessandro Sordoni, Aaron Courville, and Siva Reddy. 2025. The markovian thinker: Architecture-agnostic linear scaling of reasoning. Preprint, arXiv:2510.06557. find
  13. Moonshot AI. 2025. Introducing kimi k2 thinking. find
  14. Alon Albalak et al. 2025. Big-math: A large-scale, high-quality math dataset for reinforcement learning in language models. arXiv preprint arXiv:2502.17387. aslawliet. 2024a. Cn-k12. https://huggingface.co/datasets/aslawliet/ cn-k12. Hugging Face dataset of Chinese K-12 math problems. aslawliet. 20… ↗
  15. Mislav Balunović, Jasper Dekoninck, Ivo Petrov, Nikola Jovanović, and Martin Vechev. 2025. MathArena: Evaluating LLMs on uncontaminated math competitions. find
  16. Mislav Balunović, Jasper Dekoninck, Ivo Petrov, Nikola Jovanović, and Martin Vechev. 2025. Matharena: Evaluating llms on uncontaminated math competitions. Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmark. find
  17. Yunjie Ji, Xiaoyu Tian, Sitong Zhao, Haotian Wang, Shuaiting Chen, Yiping Peng, Han Zhao, and Xiangang Li. 2025. Am-thinking-v1: Advancing the frontier of reasoning at 32b scale. Preprint, arXiv:2505.08311. — [arXiv:2505.08311](https://arxiv.org/abs/2505.08311) find
  18. Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter,… — Language Models are Few-Shot Learners
  19. Nathan Lambert, Jacob Morrison, Valentina Pyatkin, Shengyi Huang, Hamish Ivison, Faeze Brahman, Lester James V. Miranda, Alisa Liu, Nouha Dziri, Shane Lyu, Yuling Gu, Saumya Malik, Victoria Graf, Jena D. Hwang, Jiangjiang Yang, Ronan Le Bras, Oyvind Tafjord, Chris Wilhelm, Luca Soldaini, Noah A. Smi… — Tulu 3: Pushing Frontiers in Open Language Model Post-Training
  20. Xinyun Chen, Renat Aksitov, Uri Alon, Jie Ren, Kefan Xiao, Pengcheng Yin, Sushant Prakash, Charles Sutton, Xuezhi Wang, and Denny Zhou. 2023. Universal self-consistency for large language model generation. Preprint, arXiv:2311.17311.
  21. Google DeepMind. 2025. Gemini 2.5 deep think model card. find
  22. DeepSeek. 2025. Deepseek-v3.1-terminus. find
  23. DeepSeek-AI. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. Preprint, arXiv:2501.12948.
  24. Ahmed El-Kishky, Alexander Wei, Andre Saraiva, Borys Minaiev, Daniel Selsam, David Dohan, Francis Song, Hunter Lightman, Ignasi Clavera, Jakub Pachocki, Jerry Tworek, Lorenz Kuhn, Lukasz Kaiser, Mark Chen, Max Schwarzer, Mostafa Rohaninejad, Nat McAleese, o3 contributors, Oleg Mürk, Rhythm Garg, Rui… find
  25. Etash Guha et al. 2025. Openthoughts: Data recipes for reasoning models. arXiv preprint arXiv:2506.04178.
  26. X. He et al. 2025. Deepmath-103k: A large-scale, challenging math qa benchmark. arXiv preprint arXiv:2504.11456. find
  27. Jingcheng Hu, Yinmin Zhang, Qi Han, Daxin Jiang, Xiangyu Zhang, and Heung-Yeung Shum. 2025. Open-Reasoner-Zero: An open source approach to scaling up reinforcement learning on the base model. arXiv preprint arXiv:2503.24290.
  28. Naman Jain, King Han, Alex Gu, Wen-Ding Li, Fanjia Yan, Tianjun Zhang, Sida Wang, Armando Solar-Lezama, Koushik Sen, and Ion Stoica. 2024. Livecodebench: Holistic and contamination free evaluation of large language models for code. arXiv preprint arXiv:2403.07974.
  29. Jia Li et al. 2024. Numinamath. https://huggingface.co/datasets/AI-MO/ NuminaMath-CoT. Technical report and dataset card. ↗
  30. Rongao Li, Jie Fu, Bo-Wen Zhang, Tao Huang, Zhihong Sun, Chen Lyu, Guang Liu, Zhi Jin, and Ge Li. 2023. Taco: Topics in algorithmic code generation dataset. Preprint, arXiv:2312.14852. find
  31. Yujia Li, David Choi, Junyoung Chung, Nate Kushman, Julian Schrittwieser, Rémi Leblond, Tom Eccles, James Keeling, Felix Gimeno, Agustin Dal Lago, Thomas Hubert, Peter Choy, Cyprien de Masson d’Autume, Igor Babuschkin, Xinyun Chen, Po-Sen Huang, Johannes Welbl, Sven Gowal, Alexey Cherepanov, James M… — Competition-Level Code Generation with AlphaCode
  32. LIMO Authors. 2025. Less is more for reasoning: Semi-parametric math reasoners. arXiv preprint arXiv:2502.03387.
  33. Michael Luo, Sijun Tan, Roy Huang, Ameen Patel, Alpay Ariyak, Qingyang Wu, Xiaoxiang Shi, Rachel Xin, Colin Cai, Maurice Weber, Ce Zhang, Li Erran Li, Raluca Ada Popa, and Ion Stoica. 2025. Deepcoder: A fully open-source 14b coder at o3-mini level. https://www.together.ai/blog/ deepcoder. Technical… ↗
  34. Thang Luong, Dawsen Hwang, Hoang H. Nguyen, Golnaz Ghiasi, Yuri Chervonyi, Insuk Seo, Junsu Kim, Garrett Bingham, Jonathan Lee, Swaroop Mishra, Alex Zhai, Clara Huiyi Hu, Henryk Michalewski, Jimin Kim, Jeonghyun Ahn, Junhwi Bae, Xingyou Song, Trieu H. Trinh, Quoc V. Le, and Junehyuk Jung. 2025. Towa… — Towards Robust Mathematical Reasoning
  35. Arindam Mitra, Hamed Khanpour, Corby Rosset, and Ahmed Awadallah. 2024. Orca-math: Unlocking the potential of slms in grade school math. arXiv preprint arXiv:2402.14830.
  36. Niklas Muennighoff et al. 2025. s1: Simple test-time scaling. https://arxiv.org/abs/ 2501.19393. Preprint, arXiv:2501.19393. ↗
  37. Open-R1 Team. 2025. Openr1-math-220k. https://huggingface.co/datasets/ open-r1/OpenR1-Math-220k. Open-source distilled math reasoning dataset. ↗
  38. OpenAI. 2024a. Introducing SWE-bench verified we’re releasing a human-validated subset of swe-bench that more. find
  39. OpenAI. 2024b. Learning to reason with llms. find
  40. OpenAI. 2025. Introducing gpt-5. find
  41. Bowen Peng, Jeffrey Quesnelle, Honglu Fan, and Enrico Shippole. 2023. Yarn: Efficient context window extension of large language models. Preprint, arXiv:2309.00071. find
  42. John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. 2017. Proximal policy optimization algorithms. Preprint, arXiv:1707.06347. find
  43. David Silver, Aja Huang, Chris J Maddison, Arthur Guez, Laurent Sifre, George Van Den Driessche, Julian Schrittwieser, Ioannis Antonoglou, Veda Panneershelvam, Marc Lanctot, et al. 2016. Mastering the game of go with deep neural networks and tree search. nature, 529(7587):484–489. find
  44. Junlin Wang, Jue Wang, Ben Athiwaratkun, Ce Zhang, and James Zou. 2024. Mixture-of-agents enhances large language model capabilities. Preprint, arXiv:2406.04692. find
  45. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023. Self-consistency improves chain of thought reasoning in language models. Preprint, arXiv:2203.11171.
  46. Zihan Wang, Siyao Liu, Yang Sun, Hongyan Li, and Kai Shen. 2025. Codecontests+: High-quality test case generation for competitive programming. Preprint, arXiv:2506.05817. find
  47. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. 2023. Chain-of-thought prompting elicits reasoning in large language models. Preprint, arXiv:2201.11903.
  48. Hao Wen, Yifan Su, Feifei Zhang, Yunxin Liu, Yunhao Liu, Ya-Qin Zhang, and Yuanchun Li. 2025. Parathinker: Native parallel thinking as a new paradigm to scale llm test-time compute. Preprint, arXiv:2509.04475. — [arXiv:2509.04475](https://arxiv.org/abs/2509.04475) find
  49. Tong Wu, Yang Liu, Jun Bai, Zixia Jia, Shuyi Zhang, Ziyong Lin, Yanting Wang, Song-Chun Zhu, and Zilong Zheng. 2025. Native parallel reasoner: Reasoning in parallelism via self-distilled reinforcement learning. Preprint, arXiv:2512.07461. find
  50. Ved Sirdeshmukh, Kaustubh Deshpande, Johannes Mols, Lifeng Jin, Ed-Yeremai Cardona, Dean Lee, Jeremy Kritz, Willow Primack, Summer Yue, and Chen Xing. 2025. Multichallenge: A realistic multi-turn conversation evaluation benchmark challenging to frontier llms. Preprint, arXiv:2501.17399. xAI. 2025. G… find
  51. HLE Team. 2025a. Humanity’s last exam. Preprint, arXiv:2501.14249.
  52. Yuchen Yan, Yongliang Shen, Yang Liu, Jin Jiang, Mengdi Zhang, Jian Shao, and Yueting Zhuang. 2025. Inftythink: Breaking the length limits of long-context reasoning in large language models. Preprint, arXiv:2503.06692.
  53. Kimi Team, Angang Du, Bofei Gao, Bowei Xing, Changjiu Jiang, Cheng Chen, Cheng Li, Chenjun Xiao, Chenzhuang Du, Chonghua Liao, et al. 2025. Kimi k1. 5: Scaling reinforcement learning with llms. arXiv preprint arXiv:2501.12599. — Kimi k1.5: Scaling Reinforcement Learning with LLMs
  54. Qwen Team. 2025b. Qwen3-235b-a22b-thinking-2507. find
  55. Trieu H Trinh, Yuhuai Wu, Quoc V Le, He He, and Thang Luong. 2024. Solving olympiad geometry without human demonstrations. Nature, 625(7995):476–482.
  56. Siddarth Venkatraman, Vineet Jain, Sarthak Mittal, Vedant Shah, Johan Obando-Ceron, Yoshua Bengio, Brian R. Bartoldson, Bhavya Kailkhura, Guillaume Lajoie, Glen Berseth, Nikolay Malkin, and Moksh Jain. 2025. Recursive self-aggregation unlocks deep thinking in large language models. Preprint, arXiv:2… — [arXiv:2509.26626](https://arxiv.org/abs/2509.26626) find
  57. Chunqiu Steven Xia, Yinlin Deng, Soren Dunn, and Lingming Zhang. 2024. Agentless: Demystifying llm-based software engineering agents. Preprint, arXiv:2407.01489. find
  58. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jing Zhou, Jin… — Qwen3 Technical Report
  59. Feng Yao, Liyuan Liu, Dinghuai Zhang, Chengyu Dong, Jingbo Shang, and Jianfeng Gao. 2025. Your efficient rl framework secretly brings you off-policy rl training. find
  60. In this Appendix, we provide more elaboration on the implementation details, experiment results, and qualitative results. find
  61. Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Tiantian Fan, Gaohong Liu, Lingjun Liu, Xin Liu, Haibin Lin, Zhiqi Lin, Bole Ma, Guangming Sheng, Yuxuan Tong, Chi Zhang, Mofan Zhang, Wang Zhang, Hang Zhu, Jinhua Zhu, Jiaze Chen, Jiangjie Chen, Chengyi Wang, Hongli Yu, Weinan D… — DAPO: An Open-Source LLM Reinforcement Learning System at Scale
  62. A Z.ai. 2025. Glm-4.6: Advanced agentic, reasoning and coding capabilities. find
  63. Wenting Zhao, Pranjal Aggarwal, Swarnadeep Saha, Asli Celikyilmaz, Jason Weston, and Ilia Kulikov. 2025. The majority is not always right: Rl training for solution aggregation. Preprint, arXiv:2509.06870. — [arXiv:2509.06870](https://arxiv.org/abs/2509.06870) find
  64. Tong Zheng, Hongming Zhang, Wenhao Yu, Xiaoyang Wang, Runpeng Dai, Rui Liu, Huiwen Bao, Chengsong Huang, Heng Huang, and Dong Yu. 2025. Parallel-r1: Towards parallel thinking via reinforcement learning. Preprint, arXiv:2509.07980. — [arXiv:2509.07980](https://arxiv.org/abs/2509.07980) find
  65. Initial Checkpoint Derivation To facilitate effective test-time scaling, we first implement a reasoning-oriented post-training pipeline for Qwen3-8B-Base, comprising Supervised Fine-Tuning (SFT) and Reinforcement Learning with Verifiable Rewards (RLVR). find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/hu-2026-pacore-learning-scale-test-time/