Elamur Research⌕
☰
home / papers / 2026 / Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning

Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning

Paper · 2026 · Nathanaël Carraz Rakotonirina, Ren Pang, Neha Anna John et al.

Summary

Reasoning LLMs often keep producing chain-of-thought tokens after they have already reached the right answer, re-checking it many times. This "overthinking" wastes compute and can even lower accuracy. The authors propose a two-stage recipe for adaptive reasoning.

First, supervised fine-tuning warms the model up on concise traces, built in one of two ways:

  • keeping the shortest correct sample out of several (Adaptive-Answer), or
  • deleting the summary that follows the reasoning so the model answers directly (Format-Adaptive-Answer).

Second, GRPO reinforcement learning adds a length penalty equal to the fraction of tokens emitted after the first correct answer. Revising a wrong answer is therefore not punished.

Training uses only math problems, yet evaluation spans seven benchmarks covering math, science, code, commonsense and long-context QA, on Qwen3 and DeepSeek models. Responses become 28% (8B) and 40% (32B) shorter, at a cost of 1.6–2.5 accuracy points. The method also outperforms more elaborate efficient-reasoning baselines on AUC OAA. It matters because it lowers inference cost without manually fixing a thinking budget per input.

Key points

  • Training data: a 13K-problem subset of DeepScaleR, which draws on AIME 1983–2023, AMC, Omni-Math and STILL.
  • Reward: correctness/format reward minus λ·(share of tokens after the first correct answer), with λ = 1. Rejection sampling draws 8 continuations per problem.
  • Models and benchmarks:
    • Models: Qwen3-1.7B, Qwen3-8B (main), Qwen3-32B and DeepSeek-R1-Qwen-7B-distilled.
    • Benchmarks: MATH-500, AIME 24, AIME 25, GPQA Diamond, CommonsenseQA, LiveCodeBench (v6) and LongBenchv2.
  • AUC OAA result: 76.6, beating the base model by 5 points and the second-best method by 2.5.
  • Qualitative effect: on an AIME 24 example, the base model re-verifies a correct answer seven times, Adaptive-Answer twice and Format-Adaptive-Answer not at all.

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

🎓Universitat Pompeu FabraBarcelona · Spain47 works · 11,723 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 Nathanaël Carraz Rakotonirina, Ren Pang, Neha Anna John, Michael Bohlke-Schneider and Momchil Hardalov. 2026. Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning. 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.622

@inproceedings{rakotonirina2026correct,
  title = {{Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning}},
  author = {Nathanaël Carraz Rakotonirina, Ren Pang, Neha Anna John, Michael Bohlke-Schneider and Momchil Hardalov},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.622},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.622}
}

References

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

  1. Yushi Bai, Shangqing Tu, Jiajie Zhang, Hao Peng, Xiaozhi Wang, Xin Lv, Shulin Cao, Jiazheng Xu, Lei Hou, Yuxiao Dong, Jie Tang, and Juanzi Li. 2025. LongBench v2: Towards Deeper Understanding and Reasoning on Realistic Long-Context Multitasks. In Proceedings of the 63rd Annual Meeting of the Associa…
  2. Xingyu Chen, Jiahao Xu, Tian Liang, Zhiwei He, Jianhui Pang, Dian Yu, Linfeng Song, Qiuzhi Liu, Mengfei Zhou, Zhuosheng Zhang, Rui Wang, Zhaopeng Tu, Haitao Mi, and Dong Yu. 2025. Do NOT Think That Much for 2+3=? On the Overthinking of Long Reasoning Models. In Proceedings of the 42nd International… find
  3. Yingqian Cui, Pengfei He, Jingying Zeng, Hui Liu, Xianfeng Tang, Zhenwei Dai, Yan Han, Chen Luo, Jing Huang, Zhen Li, Suhang Wang, Yue Xing, Jiliang Tang, and Qi He. 2025. Stepwise Perplexity-Guided Refinement for Efficient Chain-of-Thought Reasoning in Large Language Models. In Findings of the Asso… find
  4. DeepSeek-AI. 2025. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning. arXiv preprint arXiv:2501.12948.
  5. Bofei Gao, Feifan Song, Zhe Yang, Zefan Cai, Yibo Miao, Qingxiu Dong, Lei Li, Chenghao Ma, Liang Chen, Runxin Xu, Zhengyang Tang, Benyou Wang, Daoguang Zan, Shanghaoran Quan, Ge Zhang, Lei Sha, Yichang Zhang, Xuancheng Ren, Tianyu Liu, and Baobao Chang. 2025. Omni-MATH: A Universal Olympiad Level Ma…
  6. 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. Transactions on Machine Learning Research.
  7. Naman Jain, King Han, Alex Gu, Wen-Ding Li, Fanjia Yan, Tianjun Zhang, Sida Wang, Armando Solar-Lezama, Koushik Sen, and Ion Stoica. 2025. Live-CodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code. In The Thirteenth International Conference on Learning Representati… find
  8. Aviral Kumar, Vincent Zhuang, Rishabh Agarwal, Yi Su, John D. Co-Reyes, Avi Singh, Kate Baumli, Shariq Iqbal, Colton Bishop, Rebecca Roelofs, Lei M. Zhang, Kay McKinney, Disha Shrivastava, Cosmin Paduraru, George Tucker, Doina Precup, Feryal Behbahani, and Aleksandra Faust. 2025. Training Language M…
  9. Hunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. 2024. Let’s Verify Step by Step. In The Twelfth International Conference on Learning Representations, ICLR ’24, Vienna, Austria.
  10. Wei Liu, Ruochen Zhou, Yiyun Deng, Yuzhen Huang, Junteng Liu, Yuntian Deng, Yizhe Zhang, and Junxian He. 2025. Learn to Reason Efficiently with Adaptive Length-Based Reward Shaping. arXiv preprint arXiv:2505.15612. — [arXiv:2505.15612](https://arxiv.org/abs/2505.15612) find
  11. 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
  12. Michael Luo, Sijun Tan, Justin Wong, Xiaoxiang Shi, William Y. Tang, Manan Roongta, Colin Cai, Jeffrey Luo, Li Erran Li, Raluca Ada Popa, and Ion Stoica. 2025. DeepScaleR: Surpassing O1-Preview with a 1.5B Model by Scaling RL. Notion Blog. find
  13. Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, Shashank Gupta, Bodhisattwa Prasad Majumder, Katherine Hermann, Sean Welleck, Amir Yazdanbakhsh, and Peter Clark. 2023. Self-Refine: Iterative Refinement with… find
  14. Yingqian Min, Zhipeng Chen, Jinhao Jiang, Jie Chen, Jia Deng, Yiwen Hu, Yiru Tang, Jiapeng Wang, Xiaoxue Cheng, Huatong Song, and 1 others. 2024. Imitate, Explore, and Self-Improve: A Reproduction Report on Slow-Thinking Reasoning Systems. arXiv preprint arXiv:2412.09413. find
  15. 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. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, EMNLP ’…
  16. OpenAI. 2024. OpenAI o1 System Card. arXiv preprint arXiv:2412.16720.
  17. OpenAI. 2025. OpenAI o3 and o4-mini System Card. https://cdn.openai.com/pdf/ 2221c875-02dc-4789-800b-e7758f3722c1/ o3-and-o4-mini-system-card.pdf. ↗
  18. 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 Proceedings of the First Conference on Language Modeling, COLM ’24, Philadelphia, Pennsylvania, USA.
  19. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, Y. K. Li, Yang Wu, and 1 others. 2024. DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models. arXiv preprint arXiv:2402.03300.
  20. Guangming Sheng, Chi Zhang, Zilingfeng Ye, Xibin Wu, Wang Zhang, Ru Zhang, Yanghua Peng, Haibin Lin, and Chuan Wu. 2025. HybridFlow: A Flexible and Efficient RLHF Framework. In Proceedings of the Twentieth European Conference on Computer Systems, EuroSys ’25, pages 1279–1297, Rotterdam, Netherlands.
  21. Charlie Victor Snell, Jaehoon Lee, Kelvin Xu, and Aviral Kumar. 2025. Scaling LLM Test-Time Compute Optimally Can be More Effective than Scaling Parameters for Reasoning. In The Thirteenth International Conference on Learning Representations, ICLR ’25, Singapore, Singapore. find
  22. 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,…
  23. 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.
  24. Songjun Tu, Jiahao Lin, Qichao Zhang, Xiangyu Tian, Linjing Li, Xiangyuan Lan, and Dongbin Zhao. 2025. Learning When to Think: Shaping Adaptive Reasoning in R1-Style Models via Multi-Stage RL. In The Thirty-Ninth Annual Conference on Neural Information Processing Systems, NeurIPS ’25, San Diego, Cal… find
  25. 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, ICLR ’23, Kigali, Rwanda.
  26. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc V. Le, and Denny Zhou. 2022. Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. In Advances in Neural Information Processing Systems 35, volume 35 of NeurIPS ’22, New Orleans, Louisiana…
  27. Yuyang Wu, Yifei Wang, Ziyu Ye, Tianqi Du, Stefanie Jegelka, and Yisen Wang. 2026. When More is Less: Understanding Chain-of-Thought Length in LLMs. In The Fourteenth International Conference on Learning Representations, ICLR ’26, Kigali, Rwanda. find
  28. Heming Xia, Chak Tou Leong, Wenjie Wang, Yongqi Li, and Wenjie Li. 2025. TokenSkip: Controllable Chain-of-Thought Compression in LLMs. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, EMNLP ’25, pages 3351–3363, Suzhou, China.
  29. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 2025a. Qwen3 Technical Report. arXiv preprint arXiv:2505.09388.
  30. Junjie Yang, Ke Lin, and Xing Yu. 2025b. Think When You Need: Self-Adaptive Chain-of-Thought Learning. arXiv preprint arXiv:2504.03234. — [arXiv:2504.03234](https://arxiv.org/abs/2504.03234) find
  31. Wenkai Yang, Shuming Ma, Yankai Lin, and Furu Wei. 2025c. Towards Thinking-Optimal Scaling of Test-Time Compute for LLM Reasoning. In The Thirty-Ninth Annual Conference on Neural Information Processing Systems, NeurIPS ’25, San Diego, California, USA.
  32. Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Weinan Dai, Tiantian Fan, Gaohong Liu, Lingjun Liu, and 1 others. 2025. DAPO: An Open-Source LLM Reinforcement Learning System at Scale. arXiv preprint arXiv:2503.14476.
  33. Jiajie Zhang, Nianyi Lin, Lei Hou, Ling Feng, and Juanzi Li. 2025. AdaptThink: Reasoning Models Can Learn When to Think. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, EMNLP ’25, pages 3696– 3715, Suzhou, China.
  34. 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/rakotonirina-2026-correct-concise-complete-multi-stage/