SmartThinker: Progressive Chain-of-Thought Length Calibration for Efficient Large Language Model Reasoning

Chenzhi Hu, Qinzhe Hu, Yuhang Xu, Junyi Chen, Ruijie Wang, Shengzhong Liu, Jianxin Li, Fan Wu, Guihai Chen
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:45111-45128, 2026.

Abstract

Large reasoning models (LRMs) like OpenAI o1 and DeepSeek-R1 achieve high accuracy on complex tasks by adopting long chain-of-thought (CoT) reasoning paths. However, the inherent verbosity of these processes frequently results in redundancy and overthinking. To address this issue, existing works leverage Group Relative Policy Optimization (GRPO) to reduce LRM output length, but their static length-reward designs fail to adapt to problem difficulty and response-length distributions, causing over-compression and compromised accuracy. Therefore, we propose SmartThinker, a novel GRPO-based efficient reasoning method with progressive CoT length calibration. SmartThinker makes a two-fold contribution: First, it dynamically estimates the optimal length with peak accuracy during training and guides overlong responses toward it to reduce reasoning length while sustaining accuracy. Second, it dynamically modulates the length-reward coefficient to avoid the unwarranted penalization of correct reasoning paths. Extensive experimental results show that SmartThinker achieves up to 52.6% length compression with improved accuracy and achieves up to 16.6% accuracy relative improvement on challenging benchmarks like AIME25. The source code can be found at https://github.com/SJTU-RTEAS/SmartThinker.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-hu26v, title = {{S}mart{T}hinker: Progressive Chain-of-Thought Length Calibration for Efficient Large Language Model Reasoning}, author = {Hu, Chenzhi and Hu, Qinzhe and Xu, Yuhang and Chen, Junyi and Wang, Ruijie and Liu, Shengzhong and Li, Jianxin and Wu, Fan and Chen, Guihai}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {45111--45128}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/hu26v/hu26v.pdf}, url = {https://proceedings.mlr.press/v306/hu26v.html}, abstract = {Large reasoning models (LRMs) like OpenAI o1 and DeepSeek-R1 achieve high accuracy on complex tasks by adopting long chain-of-thought (CoT) reasoning paths. However, the inherent verbosity of these processes frequently results in redundancy and overthinking. To address this issue, existing works leverage Group Relative Policy Optimization (GRPO) to reduce LRM output length, but their static length-reward designs fail to adapt to problem difficulty and response-length distributions, causing over-compression and compromised accuracy. Therefore, we propose SmartThinker, a novel GRPO-based efficient reasoning method with progressive CoT length calibration. SmartThinker makes a two-fold contribution: First, it dynamically estimates the optimal length with peak accuracy during training and guides overlong responses toward it to reduce reasoning length while sustaining accuracy. Second, it dynamically modulates the length-reward coefficient to avoid the unwarranted penalization of correct reasoning paths. Extensive experimental results show that SmartThinker achieves up to 52.6% length compression with improved accuracy and achieves up to 16.6% accuracy relative improvement on challenging benchmarks like AIME25. The source code can be found at https://github.com/SJTU-RTEAS/SmartThinker.} }
Endnote
%0 Conference Paper %T SmartThinker: Progressive Chain-of-Thought Length Calibration for Efficient Large Language Model Reasoning %A Chenzhi Hu %A Qinzhe Hu %A Yuhang Xu %A Junyi Chen %A Ruijie Wang %A Shengzhong Liu %A Jianxin Li %A Fan Wu %A Guihai Chen %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-hu26v %I PMLR %P 45111--45128 %U https://proceedings.mlr.press/v306/hu26v.html %V 306 %X Large reasoning models (LRMs) like OpenAI o1 and DeepSeek-R1 achieve high accuracy on complex tasks by adopting long chain-of-thought (CoT) reasoning paths. However, the inherent verbosity of these processes frequently results in redundancy and overthinking. To address this issue, existing works leverage Group Relative Policy Optimization (GRPO) to reduce LRM output length, but their static length-reward designs fail to adapt to problem difficulty and response-length distributions, causing over-compression and compromised accuracy. Therefore, we propose SmartThinker, a novel GRPO-based efficient reasoning method with progressive CoT length calibration. SmartThinker makes a two-fold contribution: First, it dynamically estimates the optimal length with peak accuracy during training and guides overlong responses toward it to reduce reasoning length while sustaining accuracy. Second, it dynamically modulates the length-reward coefficient to avoid the unwarranted penalization of correct reasoning paths. Extensive experimental results show that SmartThinker achieves up to 52.6% length compression with improved accuracy and achieves up to 16.6% accuracy relative improvement on challenging benchmarks like AIME25. The source code can be found at https://github.com/SJTU-RTEAS/SmartThinker.
APA
Hu, C., Hu, Q., Xu, Y., Chen, J., Wang, R., Liu, S., Li, J., Wu, F. & Chen, G.. (2026). SmartThinker: Progressive Chain-of-Thought Length Calibration for Efficient Large Language Model Reasoning. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:45111-45128 Available from https://proceedings.mlr.press/v306/hu26v.html.

Related Material