---
title: "Efficient Reasoning for LLMs through Speculative Chain-of-Thought"
description: "Reasoning LLMs such as Deepseek-R1 are slow because large models produce very long chains of… 36 references, cited by 6 papers here."
authors: ["Jikai Wang", "Juntao Li", "Jianye Hou", "Yan Bowen", "Lijun Wu", "Min Zhang"]
year: 2026
venue: "Findings of the Association for Computational Linguistics: ACL 2026"
source: "https://aclanthology.org/2026.findings-acl.76/"
doi: "10.18653/v1/2026.findings-acl.76"
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/wang-2026-efficient-reasoning-llms-through-speculative/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/wang-2026-efficient-reasoning-llms-through-speculative/)

[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/3/) / Efficient Reasoning for LLMs through Speculative Chain-of-Thought

# Efficient Reasoning for LLMs through Speculative Chain-of-Thought

**Authors:** [Jikai Wang](https://md.research.elamur.ai/free/en/authors/jikai-wang/), [Juntao Li](https://md.research.elamur.ai/free/en/authors/juntao-li/), Jianye Hou, Yan Bowen, [Lijun Wu](https://md.research.elamur.ai/free/en/authors/lijun-wu/) and [Min Zhang](https://md.research.elamur.ai/free/en/authors/min-zhang/) · **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.76/) · [PDF](https://aclanthology.org/2026.findings-acl.76.pdf) · [DOI 10.18653/v1/2026.findings-acl.76](https://doi.org/10.18653/v1/2026.findings-acl.76) · [OpenAlex](https://openalex.org/W4416892671)

**Citations:** 0 citations (OpenAlex, 2026-10-14) · [6](https://md.research.elamur.ai/free/en/papers/wang-2026-efficient-reasoning-llms-through-speculative/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
Reasoning LLMs such as Deepseek-R1 are slow because large models produce very long chains of thought (CoT). Earlier speed-ups either shrink the model or shorten the CoT, and both can hurt accuracy on hard problems. Speculative Chain-of-Thought (SCoT) instead speeds up CoT generation by pairing a small and a large model. Borrowing from speculative decoding, it works at the level of whole thoughts rather than tokens. A lightweight draft model samples several CoT drafts in parallel. The target model then picks the best draft in one forward pass, or flags that every draft is wrong and reasons on its own. Two LoRA-based tuning steps support this. The draft model is aligned to the target model's more concise thinking style, and the target model is taught to select drafts and detect errors. On five math benchmarks, SCoT cuts latency substantially while staying close to the target model's accuracy. The approach adapts to difficulty: easy questions are handled mostly by the small model, and hard ones go to the large model.

## Key points
- **Latency results:** with Deepseek-R1-Distill-Qwen-32B (draft model: Qwen-1.5B), latency dropped by 48–66%. With Deepseek-R1-Distill-Llama-70B (draft model: 8B), it dropped by 21–49%. Average speed-ups were 2.3× and 1.6×, with a peak of 2.92×.
- **Benchmarks and baselines:** evaluated on GSM8K, MATH, GaoKao, CollegeMath and Olympiad. SCoT was more accurate than TokenSkip at comparable speed-ups. Using the same draft models, its throughput beat standard speculative decoding, by up to 2.96×.
- **Training data:**
  - Draft alignment used 1,500 target-generated CoTs from the GSM8K training set. This was needed because the 1.5B model runs about 6× faster but writes CoTs more than 4× longer.
  - Selection tuning used 500 GSM8K samples.
- **Selection accuracy:** on the GSM8K test set, the target model picked the right draft 85.1% of the time when a correct draft existed. When all drafts were wrong, it correctly flagged this 52% of the time, versus 0% before tuning.
- **Settings and limitations:** five draft chains gave the best speed–accuracy balance. The method was tested only on math, and it depends on having a suitably small draft model from the same family.

*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

- [Shanghai Artificial Intelligence Laboratory](https://md.research.elamur.ai/free/en/labs/shanghai-artificial-intelligence-laboratory/) — Research institute, Shanghai, China · 117 works · 11,281 citations
- [Soochow University](https://md.research.elamur.ai/free/en/universities/soochow-university/) — University, Suzhou, China · 63 works · 4,507 citations
- [Chinese University of Hong Kong, Shenzhen](https://md.research.elamur.ai/free/en/universities/chinese-university-of-hong-kong-shenzhen/) — University, Shenzhen, China · 47 works · 2,596 citations

📚 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: Jikai Wang, Juntao Li, Jianye Hou, Yan Bowen, Lijun Wu and Min Zhang. 2026. Efficient Reasoning for LLMs through Speculative Chain-of-Thought. 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{wang2026efficient,
  title = {{Efficient Reasoning for LLMs through Speculative Chain-of-Thought}},
  author = {Jikai Wang, Juntao Li, Jianye Hou, Yan Bowen, Lijun Wu and Min Zhang},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.76},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.76}
}
```

- [← previous Early-Exit and Instant Confidence Translation Quality Estimation](https://md.research.elamur.ai/free/en/papers/zouhar-2026-early-exit-instant-confidence-translation/)
- [next → EmoMAS: Emotion-Aware Multi-Agent System for High-Stakes Edge-Deployable Negotiation with Bayesian Orchestration](https://md.research.elamur.ai/free/en/papers/long-2026-emomas-emotion-aware-multi-agent/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/3/)

## References

36 references, 18 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Simon A Aytes, Jinheon Baek, and Sung Ju Hwang. 2025. [Sketch-of-thought: Efficient llm reasoning with adaptive cognitive-inspired sketching](https://md.research.elamur.ai/free/en/papers/aytes-2025-sketch-thought-efficient-llm-reasoning/). arXiv preprint arXiv:2503.05179.
2. Tianle Cai, Yuhong Li, Zhengyang Geng, Hongwu Peng, Jason D Lee, Deming Chen, and Tri Dao. 2024. [Medusa: Simple llm inference acceleration framework with multiple decoding heads](https://md.research.elamur.ai/free/en/papers/cai-2024-medusa-simple-llm-inference-acceleration/). arXiv preprint arXiv:2401.10774.
3. Charlie Chen, Sebastian Borgeaud, Geoffrey Irving, Jean-Baptiste Lespiau, Laurent Sifre, and John Jumper. 2023a. Accelerating large language model decoding with speculative sampling. arXiv preprint arXiv:2302.01318. [find](https://scholar.google.com/scholar?q=Charlie+Chen%2C+Sebastian+Borgeaud%2C+Geoffrey+Irving%2C+Jean%2DBaptiste+Lespiau%2C+Laurent+Sifre%2C+and+John+Jumper%2E+2023a%2E+Accelerating+large+language+model+decoding+with+speculative+sampling%2E+arXiv+preprint+ar)
4. Ziyi Chen, Xiaocong Yang, Jiacheng Lin, Chenkai Sun, Jie Huang, and Kevin Chen-Chuan Chang. 2023b. Cascade speculative drafting for even faster llm inference. arXiv preprint arXiv:2312.11462. — [arXiv:2312.11462](https://arxiv.org/abs/2312.11462) [find](https://scholar.google.com/scholar?q=Ziyi+Chen%2C+Xiaocong+Yang%2C+Jiacheng+Lin%2C+Chenkai+Sun%2C+Jie+Huang%2C+and+Kevin+Chen%2DChuan+Chang%2E+2023b%2E+Cascade+speculative+drafting+for+even+faster+llm+inference%2E+arXiv+preprint+arXiv%3A2312%2E11462%2E+%E2%80%94+%5BarXiv)
5. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. 2021. [Training verifiers to solve math word problems](https://md.research.elamur.ai/free/en/papers/cobbe-2021-training-verifiers-solve-math-word/). arXiv preprint arXiv:2110.14168.
6. DeepSeek-AI, Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, and 181 others. 2025. [Deepseek-r1: Incentivizing reasoning capability in ll…](https://md.research.elamur.ai/free/en/papers/deepseek-ai-2025-deepseek-r1-incentivizing-reasoning-capability/)
7. Mengru Ding, Hanmeng Liu, Zhizhang Fu, Jian Song, Wenbo Xie, and Yue Zhang. 2024. Break the chain: Large language models can be shortcut reasoners. Preprint, arXiv:2406.06580. — [arXiv:2406.06580](https://arxiv.org/abs/2406.06580) [find](https://scholar.google.com/scholar?q=Mengru+Ding%2C+Hanmeng+Liu%2C+Zhizhang+Fu%2C+Jian+Song%2C+Wenbo+Xie%2C+and+Yue+Zhang%2E+2024%2E+Break+the+chain%3A+Large+language+models+can+be+shortcut+reasoners%2E+Preprint%2C+arXiv%3A2406%2E06580%2E+%E2%80%94+%5BarXiv%3A2406%2E06580%5D%28htt)
8. 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. Preprint, arXiv:2412.06769. [find](https://scholar.google.com/scholar?q=Shibo+Hao%2C+Sainbayar+Sukhbaatar%2C+DiJia+Su%2C+Xian+Li%2C+Zhiting+Hu%2C+Jason+Weston%2C+and+Yuandong+Tian%2E+2024%2E+Training+large+language+models+to+reason+in+a+continuous+latent+space%2E+Preprint%2C+arXiv%3A2412%2E06769)
9. Chaoqun He, Renjie Luo, Yuzhuo Bai, Shengding Hu, Zhen Thai, Junhao Shen, Jinyi Hu, Xu Han, Yujie Huang, Yuxiang Zhang, Jie Liu, Lei Qi, Zhiyuan Liu, and Maosong Sun. 2024. [OlympiadBench: A challenging benchmark for promoting AGI with olympiad-level bilingual multimodal scientific problems](https://md.research.elamur.ai/free/en/papers/he-2024-olympiadbench-challenging-benchmark-promoting-agi/). In Proce…
10. Zhenyu He, Zexuan Zhong, Tianle Cai, Jason D Lee, and Di He. 2023. Rest: Retrieval-based speculative decoding. arXiv preprint arXiv:2311.08252. [find](https://scholar.google.com/scholar?q=Zhenyu+He%2C+Zexuan+Zhong%2C+Tianle+Cai%2C+Jason+D+Lee%2C+and+Di+He%2E+2023%2E+Rest%3A+Retrieval%2Dbased+speculative+decoding%2E+arXiv+preprint+arXiv%3A2311%2E08252%2E)
11. 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](https://md.research.elamur.ai/free/en/papers/hendrycks-2021-measuring-mathematical-problem-solving-math/). In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2).
12. 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](https://md.research.elamur.ai/free/en/papers/hou-2025-thinkprune-pruning-long-chain-thought/). Preprint, arXiv:2504.01296.
13. Daniele Paliotta, Junxiong Wang, Matteo Pagliardini, Kevin Y. Li, Aviv Bick, J. Zico Kolter, Albert Gu, François Fleuret, and Tri Dao. 2025. Thinking slow, fast: Scaling inference compute with distilled reasoners. Preprint, arXiv:2502.20339. — [arXiv:2502.20339](https://arxiv.org/abs/2502.20339) [find](https://scholar.google.com/scholar?q=Daniele+Paliotta%2C+Junxiong+Wang%2C+Matteo+Pagliardini%2C+Kevin+Y%2E+Li%2C+Aviv+Bick%2C+J%2E+Zico+Kolter%2C+Albert+Gu%2C+Fran%C3%A7ois+Fleuret%2C+and+Tri+Dao%2E+2025%2E+Thinking+slow%2C+fast%3A+Scaling+inference+compute+with+distill)
14. Edward J Hu, yelong shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. [LoRA: Low-rank adaptation of large language models](https://md.research.elamur.ai/free/en/papers/hu-2021-lora-low-rank-adaptation-large/). In International Conference on Learning Representations.
15. Zhuoshi Pan, Qianhui Wu, Huiqiang Jiang, Menglin Xia, Xufang Luo, Jue Zhang, Qingwei Lin, Victor Rühle, Yuqing Yang, Chin-Yew Lin, and 1 others. 2024. [Llmlingua-2: Data distillation for efficient and faithful task-agnostic prompt compression](https://md.research.elamur.ai/free/en/papers/pan-2024-llmlingua-2-data-distillation-efficient/). In Findings of the Association for Computational Linguist…
16. 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](https://md.research.elamur.ai/free/en/papers/openai-2024-openai-o1-system-card/). arXiv preprint arXiv:2412.16720.
17. Sehoon Kim, Karttikeya Mangalam, Suhong Moon, Jitendra Malik, Michael W. Mahoney, Amir Gholami, and Kurt Keutzer. 2023. Speculative decoding with big little decoder. In Thirty-seventh Conference on Neural Information Processing Systems. [find](https://scholar.google.com/scholar?q=Sehoon+Kim%2C+Karttikeya+Mangalam%2C+Suhong+Moon%2C+Jitendra+Malik%2C+Michael+W%2E+Mahoney%2C+Amir+Gholami%2C+and+Kurt+Keutzer%2E+2023%2E+Speculative+decoding+with+big+little+decoder%2E+In+Thirty%2Dseventh+Conference+on+Ne)
18. Yaniv Leviathan, Matan Kalman, and Yossi Matias. 2023. Fast inference from transformers via speculative decoding. In International Conference on Machine Learning, pages 19274–19286. PMLR. [find](https://scholar.google.com/scholar?q=Yaniv+Leviathan%2C+Matan+Kalman%2C+and+Yossi+Matias%2E+2023%2E+Fast+inference+from+transformers+via+speculative+decoding%2E+In+International+Conference+on+Machine+Learning%2C+pages+19274%E2%80%9319286%2E+PMLR%2E)
19. Yuhui Li, Fangyun Wei, Chao Zhang, and Hongyang Zhang. 2024. [Eagle: Speculative sampling requires rethinking feature uncertainty](https://md.research.elamur.ai/free/en/papers/li-2024-eagle-speculative-sampling-requires-rethinking/). arXiv preprint arXiv:2401.15077.
20. Huanxuan Liao, Shizhu He, Yupu Hao, Xiang Li, Yuanzhe Zhang, Jun Zhao, and Kang Liu. 2025. [SKIntern: Internalizing symbolic knowledge for distilling better CoT capabilities into small language models](https://md.research.elamur.ai/free/en/papers/liao-2025-skintern-internalizing-symbolic-knowledge-distilling/). In Proceedings of the 31st International Conference on Computational Linguistics, pages 3203–3221,…
21. Minpeng Liao, Wei Luo, Chengxi Li, Jing Wu, and Kai Fan. 2024. Mario: Math reasoning with code interpreter output – a reproducible pipeline. Preprint, arXiv:2401.08190. [find](https://scholar.google.com/scholar?q=Minpeng+Liao%2C+Wei+Luo%2C+Chengxi+Li%2C+Jing+Wu%2C+and+Kai+Fan%2E+2024%2E+Mario%3A+Math+reasoning+with+code+interpreter+output+%E2%80%93+a+reproducible+pipeline%2E+Preprint%2C+arXiv%3A2401%2E08190%2E)
22. Ruikang Liu, Yuxuan Sun, Manyi Zhang, Haoli Bai, Xianzhi Yu, Tiezheng Yu, Chun Yuan, and Lu Hou. 2025. Quantization hurts reasoning? an empirical study on quantized reasoning models. Preprint, arXiv:2504.04823. — [arXiv:2504.04823](https://arxiv.org/abs/2504.04823) [find](https://scholar.google.com/scholar?q=Ruikang+Liu%2C+Yuxuan+Sun%2C+Manyi+Zhang%2C+Haoli+Bai%2C+Xianzhi+Yu%2C+Tiezheng+Yu%2C+Chun+Yuan%2C+and+Lu+Hou%2E+2025%2E+Quantization+hurts+reasoning%3F+an+empirical+study+on+quantized+reasoning+models%2E+Preprint%2C+arXiv%3A2)
23. Tengxiao Liu, Qipeng Guo, Xiangkun Hu, Cheng Jiayang, Yue Zhang, Xipeng Qiu, and Zheng Zhang. 2024. Can language models learn to skip steps? In The Thirty-eighth Annual Conference on Neural Information Processing Systems. [find](https://scholar.google.com/scholar?q=Tengxiao+Liu%2C+Qipeng+Guo%2C+Xiangkun+Hu%2C+Cheng+Jiayang%2C+Yue+Zhang%2C+Xipeng+Qiu%2C+and+Zheng+Zhang%2E+2024%2E+Can+language+models+learn+to+skip+steps%3F+In+The+Thirty%2Deighth+Annual+Conference+on+Neural+Informatio)
24. 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. Preprint, arXiv:2501.12570. [find](https://scholar.google.com/scholar?q=Haotian+Luo%2C+Li+Shen%2C+Haiying+He%2C+Yibo+Wang%2C+Shiwei+Liu%2C+Wei+Li%2C+Naiqiang+Tan%2C+Xiaochun+Cao%2C+and+Dacheng+Tao%2E+2025%2E+O1%2Dpruner%3A+Length%2Dharmonizing+fine%2Dtuning+for+o1%2Dlike+reasoning+pruning%2E+Preprint%2C+a)
25. Nikunj Saunshi, Nishanth Dikkala, Zhiyuan Li, Sanjiv Kumar, and Sashank J. Reddi. 2025. Reasoning with latent thoughts: On the power of looped transformers. In The Thirteenth International Conference on Learning Representations. [find](https://scholar.google.com/scholar?q=Nikunj+Saunshi%2C+Nishanth+Dikkala%2C+Zhiyuan+Li%2C+Sanjiv+Kumar%2C+and+Sashank+J%2E+Reddi%2E+2025%2E+Reasoning+with+latent+thoughts%3A+On+the+power+of+looped+transformers%2E+In+The+Thirteenth+International+Conference+)
26. Benjamin Spector and Chris Re. 2023. Accelerating llm inference with staged speculative decoding. arXiv preprint arXiv:2308.04623. — [arXiv:2308.04623](https://arxiv.org/abs/2308.04623) [find](https://scholar.google.com/scholar?q=Benjamin+Spector+and+Chris+Re%2E+2023%2E+Accelerating+llm+inference+with+staged+speculative+decoding%2E+arXiv+preprint+arXiv%3A2308%2E04623%2E+%E2%80%94+%5BarXiv%3A2308%2E04623%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2308%2E04623%29)
27. Mitchell Stern, Noam Shazeer, and Jakob Uszkoreit. 2018. Blockwise parallel decoding for deep autoregressive models. Advances in Neural Information Processing Systems, 31. [find](https://scholar.google.com/scholar?q=Mitchell+Stern%2C+Noam+Shazeer%2C+and+Jakob+Uszkoreit%2E+2018%2E+Blockwise+parallel+decoding+for+deep+autoregressive+models%2E+Advances+in+Neural+Information+Processing+Systems%2C+31%2E)
28. Zhengyang Tang, Xingxing Zhang, Benyou Wang, and Furu Wei. 2024. Mathscale: Scaling instruction tuning for mathematical reasoning. Preprint, arXiv:2403.02884. [find](https://scholar.google.com/scholar?q=Zhengyang+Tang%2C+Xingxing+Zhang%2C+Benyou+Wang%2C+and+Furu+Wei%2E+2024%2E+Mathscale%3A+Scaling+instruction+tuning+for+mathematical+reasoning%2E+Preprint%2C+arXiv%3A2403%2E02884%2E)
29. 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](https://md.research.elamur.ai/free/en/papers/wei-2022-chain-thought-prompting-elicits-reasoning/). Preprint, arXiv:2201.11903.
30. Heming Xia, Tao Ge, Peiyi Wang, Si-Qing Chen, Furu Wei, and Zhifang Sui. 2023. [Speculative decoding: Exploiting speculative execution for accelerating seq2seq generation](https://md.research.elamur.ai/free/en/papers/xia-2023-speculative-decoding-exploiting-speculative-execution/). In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 3909–3925.
31. Heming Xia, Yongqi Li, Chak Tou Leong, Wenjie Wang, and Wenjie Li. 2025. [Tokenskip: Controllable chain-of-thought compression in llms](https://md.research.elamur.ai/free/en/papers/xia-2025-tokenskip-controllable-chain-thought-compression/). Preprint, arXiv:2502.12067.
32. Silei Xu, Wenhao Xie, Lingxiao Zhao, and Pengcheng He. 2025a. Chain of draft: Thinking faster by writing less. Preprint, arXiv:2502.18600. [find](https://scholar.google.com/scholar?q=Silei+Xu%2C+Wenhao+Xie%2C+Lingxiao+Zhao%2C+and+Pengcheng+He%2E+2025a%2E+Chain+of+draft%3A+Thinking+faster+by+writing+less%2E+Preprint%2C+arXiv%3A2502%2E18600%2E)
33. Yige Xu, Xu Guo, Zhiwei Zeng, and Chunyan Miao. 2025b. [Softcot: Soft chain-of-thought for efficient reasoning with llms](https://md.research.elamur.ai/free/en/papers/xu-2025-softcot-soft-chain-thought-efficient/). Preprint, arXiv:2502.12134.
34. Ping Yu, Jing Xu, Jason Weston, and Ilia Kulikov. 2024. Distilling system 2 into system 1. Preprint, arXiv:2407.06023. [find](https://scholar.google.com/scholar?q=Ping+Yu%2C+Jing+Xu%2C+Jason+Weston%2C+and+Ilia+Kulikov%2E+2024%2E+Distilling+system+2+into+system+1%2E+Preprint%2C+arXiv%3A2407%2E06023%2E)
35. 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](https://md.research.elamur.ai/free/en/papers/zhang-2025-lightthinker-thinking-step-step-compression/). Preprint, arXiv:2502.15589.
36. Jun Zhang, Jue Wang, Huan Li, Lidan Shou, Ke Chen, Gang Chen, and Sharad Mehrotra. 2023. [Draft & verify: Lossless large language model acceleration via self-speculative decoding](https://md.research.elamur.ai/free/en/papers/zhang-2024-draft-verify-lossless-large-language/). arXiv preprint arXiv:2309.08168.

*Updated 30 October 2026*