--- title: "Beyond Token Length — paper by Wu et al. (2026)" description: "The authors propose Step Pruner (SP), an RL framework trained with GRPO. 1 reference, cited by 1 paper here." authors: ["Canhui Wu", "Qiong Cao", "Chang Li", "Zhenfang Wang", "Chao Xue", "Yuwei Fan", "Wei Xi", "Xiaodong He"] year: 2026 venue: "Findings of the Association for Computational Linguistics: ACL 2026" source: "https://aclanthology.org/2026.findings-acl.94/" doi: "10.18653/v1/2026.findings-acl.94" licence: "CC BY 4.0 (ACL Anthology)" licence_url: "https://creativecommons.org/licenses/by/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/wu-2026-beyond-token-length-step-pruner/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/wu-2026-beyond-token-length-step-pruner/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/7/) / Beyond Token Length: Step Pruner for Efficient and Accurate Reasoning in Large Language Models # Beyond Token Length: Step Pruner for Efficient and Accurate Reasoning in Large Language Models **Authors:** Canhui Wu, Qiong Cao, Chang Li, Zhenfang Wang, [Chao Xue](https://md.research.elamur.ai/free/en/authors/chao-xue/), [Yuwei Fan](https://md.research.elamur.ai/free/en/authors/yuwei-fan/), Wei Xi and [Xiaodong He](https://md.research.elamur.ai/free/en/authors/xiaodong-he/) · **Year:** 2026 · **Venue:** [Findings of the Association for Computational Linguistics: ACL 2026](https://md.research.elamur.ai/free/en/venues/findings/) ([Findings of the Association for Computational Linguistics 2026](https://md.research.elamur.ai/free/en/venues/findings/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/) **Links:** [Publisher's page](https://aclanthology.org/2026.findings-acl.94/) · [PDF](https://aclanthology.org/2026.findings-acl.94.pdf) · [DOI 10.18653/v1/2026.findings-acl.94](https://doi.org/10.18653/v1/2026.findings-acl.94) · [OpenAlex](https://openalex.org/W4414969874) **Citations:** 1 citations (OpenAlex, 2026-10-14) · 1 in our library, of the 149,439 papers whose bibliographies we read ## Summary Large Reasoning Models often "overthink," producing very long chains of thought even for trivial questions. Earlier RL fixes penalize token counts, but fewer tokens do not always mean fewer reasoning steps, and late in training models can game such penalties by dropping the reasoning altogether. The authors propose Step Pruner (SP), an RL framework trained with GRPO. SP works in four stages: - Each sampled response is split into steps, with paragraphs as the unit. - The smallest step count among correct samples serves as the target. - Correct responses that use extra steps are penalized, while wrong responses receive no step reward. - Training halts once average output length stops decreasing, which prevents hacking through step merging. On AIME24, GPQA Diamond, MATH500 and GSM8K, using distilled DeepSeek-R1 Qwen (7B, 1.5B) and Llama-8B models, SP matches or improves accuracy while sharply cutting output length. Across these settings it gives the best accuracy–length trade-off against seven baselines, including O1-Pruner, ShorterBetter and TrainEfficient. The work suggests that counting steps, rather than tokens, is a more reliable signal for making reasoning cheaper without sacrificing quality. ## Key points - SP cuts the number of tokens it uses on AIME24 by 69.7%. On MATH500 it needs about 33% of the tokens of DeepSeek-R1-Distilled-Qwen-7B at the same accuracy level (92.0% vs. 91.8%). - Averaged over benchmarks, SP uses roughly 44% of baseline tokens. The comparison covers seven baselines, among them CCoT, CoD, Cot-Valve, Model Merge, O1-Pruner, ShorterBetter and TrainEfficient. - Among the segmentation strategies tested, paragraph-based splitting gave the best balance between compression and accuracy. - An ablation removed four components: correct reward, correct optimal step, masking of wrong responses, and skipping all-wrong groups. Each removal hurt accuracy, length, or both. - Gemini 2.5 was used to label sentences in model outputs. SP-7B outputs contain more Pivotal Reasoning and calculation, and less exploration of alternatives and self-verification. On Llama-3.1-8B, SP reaches 84.8% on MATH500 with shorter outputs. *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 - [Xi'an Jiaotong University](https://md.research.elamur.ai/free/en/universities/xi-an-jiaotong-university/) — University, Xi'an, China · 22 works · 36,046 citations Also on the byline: [Future Academy](https://ror.org/0570trv98). 📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 2,869 works 🧭 Topics (2): [🧠 Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) · 3,320 works · [🧮 Reasoning](https://md.research.elamur.ai/free/en/topics/reasoning/) · 1,426 works ## Cite Cite as: Canhui Wu, Qiong Cao, Chang Li, Zhenfang Wang, Chao Xue, Yuwei Fan, Wei Xi and Xiaodong He. 2026. Beyond Token Length: Step Pruner for Efficient and Accurate Reasoning in Large Language Models. Findings of the Association for Computational Linguistics: ACL 2026. The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{wu2026beyond, title = {{Beyond Token Length: Step Pruner for Efficient and Accurate Reasoning in Large Language Models}}, author = {Canhui Wu, Qiong Cao, Chang Li, Zhenfang Wang, Chao Xue, Yuwei Fan, Wei Xi and Xiaodong He}, booktitle = {Findings of the Association for Computational Linguistics: ACL 2026}, year = {2026}, doi = {10.18653/v1/2026.findings-acl.94}, url = {https://doi.org/10.18653/v1/2026.findings-acl.94} } ``` - [← previous Beyond Outcome Verification: Verifiable Process Reward Models for Structured Reasoning](https://md.research.elamur.ai/free/en/papers/pronesti-2026-beyond-outcome-verification-verifiable-process/) - [next → Beyond Transcripts: A Renewed Perspective on Audio Chaptering](https://md.research.elamur.ai/free/en/papers/retkowski-2026-beyond-transcripts-renewed-perspective-audio/) - [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/7/) ## References 1 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Daman Arora and Andrea Zanette. 2025. Training language models to reason efficiently. Preprint, arXiv:2502.04463. \#\# Conclusion [find](https://scholar.google.com/scholar?q=Daman+Arora+and+Andrea+Zanette%2E+2025%2E+Training+language+models+to+reason+efficiently%2E+Preprint%2C+arXiv%3A2502%2E04463%2E+%5C%23%5C%23+Conclusion) *Updated 31 October 2026*