Elamur Research⌕
☰
home / papers / 2025 / ConCISE: Confidence-guided Compression in Step-by-step Efficient Reasoning

ConCISE: Confidence-guided Compression in Step-by-step Efficient Reasoning

Paper · 2025 · Ziqing Qiao, Yongheng Deng, Jiali Zeng et al.

Summary

Large Reasoning Models (LRMs) such as DeepSeek-R1 tend to produce very long chains of thought, which raises inference cost. Earlier compression approaches fine-tune models on shorter traces. Those traces are built either by choosing the shortest correct sample, which can still contain wasted steps, or by deleting steps after generation, which can break the logical flow. The authors treat reflection as something triggered when the model's internal confidence falls below a threshold. From this view they identify two sources of redundancy. Confidence Deficit is when the model re-checks steps that are already correct. Termination Delay is when it keeps reasoning after reaching a verified answer.

Their framework, ConCISE, addresses each during data generation:

  • Confidence Injection inserts confidence-boosting phrases at likely reflection points.
  • Early Stopping uses a lightweight confidence detector to end reasoning once the answer is settled.

Fine-tuning LRMs on the resulting traces with SFT or SimPO gives a better balance of length and accuracy than baselines. Under SimPO, output length drops by roughly half while accuracy stays high. This makes cheaper reasoning practical in resource-constrained settings.

Key points

  • Four mainstream LRMs were fine-tuned on ConCISE data with SFT and SimPO. Evaluation used math benchmarks including Math-500, GSM8K and AIME24, with DeepSeek-R1-Distill-Qwen-7B and 1.5B among the models.
  • The SimPO variant cuts average response length by about 50% while keeping task accuracy high.
  • An ablation on DeepSeek-R1-Distill-Qwen-7B under SFT tested each component alone. Either one keeps accuracy near the full method but compresses less: Confidence Injection mainly trims reflection before the answer, and Early Stopping mainly trims it afterwards.
  • ConCISE was compared with training-free methods TALE and DEER (λ = 0.95 and 0.98). DEER loses 5–8% accuracy, and TALE is unstable across datasets and models. Both ConCISE-SimPO and a training-free variant, ConCISE-Decoding, give better trade-offs.
  • Stated limitations:
    • Non-reflection steps could still be shortened further.
    • The method has no direct confidence estimate before the first answer.
    • Integration with RLVR methods such as GRPO is left to future work.

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🏢TencentShenzhen · China134 works · 19,743 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 2Research subjects and languages the work is about.Reasoning · 1,426 worksEfficient models & compression · 242 works

Cite

Cite as Ziqing Qiao, Yongheng Deng, Jiali Zeng, Dong Wang, Lai Wei, Guanbo Wang, Fandong Meng, Jie Zhou, Ju Ren and Yaoxue Zhang. 2025. ConCISE: Confidence-guided Compression in Step-by-step Efficient Reasoning. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.emnlp-main.405

@inproceedings{qiao2025concise,
  title = {{ConCISE: Confidence-guided Compression in Step-by-step Efficient Reasoning}},
  author = {Ziqing Qiao, Yongheng Deng, Jiali Zeng, Dong Wang, Lai Wei, Guanbo Wang, Fandong Meng, Jie Zhou, Ju Ren and Yaoxue Zhang},
  booktitle = {Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing},
  year = {2025},
  doi = {10.18653/v1/2025.emnlp-main.405},
  url = {https://doi.org/10.18653/v1/2025.emnlp-main.405}
}

References

44 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. Daman Arora and Andrea Zanette. 2025. Training language models to reason efficiently. arXiv preprint arXiv:2502.04463. find
  3. Simon A Aytes, Jinheon Baek, and Sung Ju Hwang. 2025. Sketch-of-thought: Efficient llm reasoning with adaptive cognitive-inspired sketching. arXiv preprint arXiv:2503.05179.
  4. 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.
  5. Yu-Neng Chuang, Leisheng Yu, Guanchu Wang, Lizhe Zhang, Zirui Liu, Xuanting Cai, Yang Sui, Vladimir Braverman, and Xia Hu. 2025. Confident or seek stronger: Exploring uncertainty-based on-device llm routing from benchmarking to generalization. arXiv preprint arXiv:2502.04428. — [arXiv:2502.04428](https://arxiv.org/abs/2502.04428) find
  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. Yingqian Cui, Pengfei He, Jingying Zeng, Hui Liu, Xianfeng Tang, Zhenwei Dai, Yan Han, Chen Luo, Jing Huang, Zhen Li, and 1 others. 2025. Stepwise perplexity-guided refinement for efficient chain-of-thought reasoning in large language models. arXiv preprint arXiv:2502.13260. find
  8. Sicheng Feng, Gongfan Fang, Xinyin Ma, and Xinchao Wang. 2025. Efficient reasoning models: A survey. arXiv preprint arXiv:2504.10903.
  9. 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.
  10. Tingxu Han, Zhenting Wang, Chunrong Fang, Shiyu Zhao, Shiqing Ma, and Zhenyu Chen. 2024. Token-budget-aware llm reasoning. arXiv preprint arXiv:2412.18547.
  11. Shibo Hao, Sainbayar Sukhbaatar, DiJia Su, Xian Li, Zhiting Hu, Jason Weston, and Yuandong Tian. 2024. Training large language models to reason in a continuous latent space. arXiv preprint arXiv:2412.06769. find
  12. Jujie He, Jiacai Liu, Chris Yuhao Liu, Rui Yan, Chaojie Wang, Peng Cheng, Xiaoyu Zhang, Fuxiang Zhang, Jiacheng Xu, Wei Shen, Siyuan Li, Liang Zeng, Tianwen Wei, Cheng Cheng, Bo An, Yang Liu, and Yahui Zhou. 2025. Skywork open reasoner series. https://capricious-hydrogen-41c.notion.si te/Skywork-Ope… ↗
  13. Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. 2021. Measuring mathematical problem solving with the math dataset. NeurIPS.
  14. 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.
  15. Ayeong Lee, Ethan Che, and Tianyi Peng. 2025. How well do llms compress their own chain-of-thought? a token complexity approach. arXiv preprint arXiv:2503.01141. — [arXiv:2503.01141](https://arxiv.org/abs/2503.01141) find
  16. Haotian Luo, Li Shen, Haiying He, Yibo Wang, Shiwei Liu, Wei Li, Naiqiang Tan, Xiaochun Cao, and Dacheng Tao. 2025. O1-pruner: Length-harmonizing fine-tuning for o1-like reasoning pruning. arXiv preprint arXiv:2501.12570. find
  17. Xinyin Ma, Guangnian Wan, Runpeng Yu, Gongfan Fang, and Xinchao Wang. 2025. Cot-valve: Length-compressible chain-of-thought tuning. arXiv preprint arXiv:2502.09601.
  18. Yu Meng, Mengzhou Xia, and Danqi Chen. 2024. Simpo: Simple preference optimization with a reference-free reward. Advances in Neural Information Processing Systems, 37:124198–124235. find
  19. Niklas Muennighoff, Zitong Yang, Weijia Shi, Xiang Lisa Li, Li Fei-Fei, Hannaneh Hajishirzi, Luke Zettlemoyer, Percy Liang, Emmanuel Candès, and Tatsunori Hashimoto. 2025. s1: Simple test-time scaling. arXiv preprint arXiv:2501.19393.
  20. Tergel Munkhbat, Namgyu Ho, Seo Hyun Kim, Yongjin Yang, Yujin Kim, and Se-Young Yun. 2025. Self-training elicits concise reasoning in large language models. arXiv preprint arXiv:2502.20122. find
  21. Sania Nayab, Giulio Rossolini, Marco Simoni, Andrea Saracino, Giorgio Buttazzo, Nicolamaria Manes, and Fabrizio Giacomelli. 2024. Concise thoughts: Impact of output length on llm reasoning and cost. arXiv preprint arXiv:2407.19825. — [arXiv:2407.19825](https://arxiv.org/abs/2407.19825) find
  22. Isaac Ong, Amjad Almahairi, Vincent Wu, Wei-Lin Chiang, Tianhao Wu, Joseph E Gonzalez, M Waleed Kadous, and Ion Stoica. 2024. Routellm: Learning to route llms from preference data. In The Thirteenth International Conference on Learning Representations. find
  23. Xiaoye Qu, Yafu Li, Zhaochen Su, Weigao Sun, Jianhao Yan, Dongrui Liu, Ganqu Cui, Daizong Liu, Shuxian Liang, Junxian He, and 1 others. 2025. A survey of efficient reasoning for large reasoning models: Language, multimodality, and beyond. arXiv preprint arXiv:2503.21614.
  24. Qwen Team. 2024. QwQ: Reflect Deeply on the Boundaries of the Unknown. Accessed: 2025-04-05. find
  25. Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. 2023. Direct preference optimization: Your language model is secretly a reward model. Advances in Neural Information Processing Systems, 36:53728– 53741.
  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. Matthew Renze and Erhan Guven. 2024. The benefits of a concise chain of thought on problem-solving in large language models. In 2024 2nd International Conference on Foundation and Large Language Models (FLLM), pages 476–483. IEEE. find
  28. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, and 1 others. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300.
  29. Yi Shen, Jian Zhang, Jieyun Huang, Shuming Shi, Wenjing Zhang, Jiangze Yan, Ning Wang, Kai Wang, and Shiguo Lian. 2025a. Dast: Difficulty-adaptive slow-thinking for large reasoning models. arXiv preprint arXiv:2503.04472. find
  30. Zhenyi Shen, Hanqi Yan, Linhai Zhang, Zhanghao Hu, Yali Du, and Yulan He. 2025b. Codi: Compressing chain-of-thought into continuous space via self-distillation. arXiv preprint arXiv:2502.21074.
  31. Yang Sui, Yu-Neng Chuang, Guanchu Wang, Jiamu Zhang, Tianyi Zhang, Jiayi Yuan, Hongyi Liu, Andrew Wen, Hanjie Chen, Xia Hu, and 1 others. 2025. Stop overthinking: A survey on efficient reasoning for large language models. arXiv preprint arXiv:2503.16419.
  32. 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 preprint arXiv:2410.20290. find
  33. Kimi Team, Angang Du, Bofei Gao, Bowei Xing, Changjiu Jiang, Cheng Chen, Cheng Li, Chenjun Xiao, Chenzhuang Du, Chonghua Liao, and 1 others. 2025. Kimi k1. 5: Scaling reinforcement learning with llms. arXiv preprint arXiv:2501.12599. — Kimi k1.5: Scaling Reinforcement Learning with LLMs
  34. Yue Wang, Qiuzhi Liu, Jiahao Xu, Tian Liang, Xingyu Chen, Zhiwei He, Linfeng Song, Dian Yu, Juntao Li, Zhuosheng Zhang, and 1 others. 2025. Thoughts are all over the place: On the underthinking of o1-like llms. arXiv preprint arXiv:2501.18585. find
  35. Jason Wei, Maarten Bosma, Vincent Y Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M Dai, and Quoc V Le. 2021. Finetuned language models are zero-shot learners. arXiv preprint arXiv:2109.01652.
  36. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824– 24837.
  37. Yuyang Wu, Yifei Wang, Tianqi Du, Stefanie Jegelka, and Yisen Wang. 2025. When more is less: Understanding chain-of-thought length in llms. arXiv preprint arXiv:2502.07266. find
  38. Heming Xia, Yongqi Li, Chak Tou Leong, Wenjie Wang, and Wenjie Li. 2025. Tokenskip: Controllable chain-of-thought compression in llms. arXiv preprint arXiv:2502.12067.
  39. Silei Xu, Wenhao Xie, Lingxiao Zhao, and Pengcheng He. 2025. Chain of draft: Thinking faster by writing less. arXiv preprint arXiv:2502.18600. find
  40. 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, and 22 others. 2024. Qwen2.5 technical report. arXiv preprint arXi… find
  41. Chenxu Yang, Qingyi Si, Yongjie Duan, Zheliang Zhu, Chenyu Zhu, Zheng Lin, Li Cao, and Weiping Wang. 2025. Dynamic early exit in reasoning models. arXiv preprint arXiv:2504.15895. find
  42. Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Tiantian Fan, Gaohong Liu, Lingjun Liu, Xin Liu, and 1 others. 2025. Dapo: An open-source llm reinforcement learning system at scale. arXiv preprint arXiv:2503.14476.
  43. Jintian Zhang, Yuqi Zhu, Mengshu Sun, Yujie Luo, Shuofei Qiao, Lun Du, Da Zheng, Huajun Chen, and Ningyu Zhang. 2025. Lightthinker: Thinking step-by-step compression. arXiv preprint arXiv:2502.15589.
  44. Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, Zheyan Luo, Zhangchi Feng, and Yongqiang Ma. 2024. Llamafactory: Unified efficient fine-tuning of 100+ language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations),…

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/qiao-2025-concise-confidence-guided-compression-step/