Elamur Research⌕
☰
home / papers / 2025 / Chain of Draft: Thinking Faster by Writing Less

Chain of Draft: Thinking Faster by Writing Less

Paper · 2025 · Silei Xu, Wenhao Xie, Lingxiao Zhao et al.

Summary

Chain-of-Thought prompting improves LLM reasoning, but the long step-by-step explanations it produces increase token costs and response time. The authors, from Zoom Communications, propose Chain of Draft (CoD). This prompting strategy mimics how people jot down brief notes when solving problems. The system prompt asks the model to reason step by step but to keep each step to a draft of at most five words. This is a soft guideline, not an enforced constraint. The few-shot examples include drafts written by hand. The authors compare CoD with Standard few-shot prompting and CoT on GPT-4o and Claude 3.5 Sonnet across four tasks: GSM8k for arithmetic, BIG-bench date and sports understanding for commonsense, and a synthetic coin-flip set for symbolic reasoning. CoD usually matches or beats CoT accuracy while using far fewer tokens, sometimes as little as 7.6% of CoT's output, which also reduces latency. However, CoD's advantage weakens in zero-shot use and on small models. The authors attribute this to the absence of draft-style reasoning in training data. The work matters for latency- and budget-sensitive deployments.

Key points

  • Sports understanding: On Claude 3.5 Sonnet, CoD cut average output from 189.4 to 14.3 tokens, a 92.4% reduction. Accuracy rose from 93.2% to 97.3%, and latency fell from 3.6 s to 1.0 s.
  • Date understanding: On GPT-4o, CoD scored 88.1% with 30.2 tokens, versus 90.2% with 75.7 tokens for CoT.
  • Coin flip: The authors built a 250-example test set using names from NameDataset. Both models reached 100% with CoT and with CoD, and CoD used 68–86% fewer tokens.
  • Zero-shot GSM8K: CoD falls well short of CoT. On Claude 3.5 Sonnet it scored 65.5% versus 90.4% for CoT, only 3.6 points above Standard prompting.
  • Small models (under 3B parameters): On Qwen2.5 1.5B/3B, Llama 3.2 3B and Zoom SLM 2.3B, CoD beats direct answering but trails CoT, for example 43.1% versus 59.1% on Qwen2.5-3B. The code is available at github.com/sileix/chain-of-draft.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksComputer Vision and Pattern Recognition · 3,486 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,685 works

Cite

Cite as Silei Xu, Wenhao Xie, Lingxiao Zhao and Pengcheng He. 2025. Chain of Draft: Thinking Faster by Writing Less. arXiv:2502.18600.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2502.18600 · arXiv 2502.18600

@misc{xu2025chain,
  title = {{Chain of Draft: Thinking Faster by Writing Less}},
  author = {Silei Xu, Wenhao Xie, Lingxiao Zhao and Pengcheng He},
  year = {2025},
  doi = {10.48550/arxiv.2502.18600},
  eprint = {2502.18600},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2502.18600}
}

References

28 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. BIG bench authors. 2023. Beyond the imitation game: Quantifying and extrapolating the capabilities of language models. Transactions on Machine Learning Research.
  2. Maciej Besta, Nils Blach, Ales Kubicek, Robert Gerstenberger, Michal Podstawski, Lukas Gianinazzi, Joanna Gajda, Tomasz Lehmann, Hubert Niewiadomski, Piotr Nyczyk, et al. 2024. Graph of thoughts: Solving elaborate problems with large language models. In Proceedings of the AAAI Conference on Artifici… find
  3. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, Chri… — Language Models are Few-Shot Learners
  4. Sijia Chen, Baochun Li, and Di Niu. 2024a. Boosting of thoughts: Trial-and-error problem solving with large language models. arXiv preprint arXiv:2402.11140. — [arXiv:2402.11140](https://arxiv.org/abs/2402.11140) find
  5. Xingyu Chen, Jiahao Xu, Tian Liang, Zhiwei He, Jianhui Pang, Dian Yu, Linfeng Song, Qiuzhi Liu, Mengfei Zhou, Zhuosheng Zhang, et al. 2024b. Do not think that much for 2+ 3=? on the overthinking of o1-like llms. arXiv preprint arXiv:2412.21187.
  6. Cheng-Han Chiang and Hung-yi Lee. 2024. Over-reasoning and redundant calculation of large language models. arXiv preprint arXiv:2401.11467.
  7. 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. arXiv preprint arXiv:2110.14168.
  8. Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783.
  9. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948.
  10. Tingxu Han, Chunrong Fang, Shiyu Zhao, Shiqing Ma, Zhenyu Chen, and Zhenting Wang. 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. Song Jiang, Zahra Shakeri, Aaron Chan, Maziar Sanjabi, Hamed Firooz, Yinglong Xia, Bugra Akyildiz, Yizhou Sun, Jinchao Li, Qifan Wang, et al. 2023. Resprompt: Residual connection prompting advances multi-step reasoning in large language models. arXiv preprint arXiv:2310.04743. find
  13. Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2022. Large language models are zero-shot reasoners. Advances in neural information processing systems, 35:22199– 22213.
  14. Bin Lei, Chunhua Liao, Caiwen Ding, et al. 2023. Boosting logical reasoning in large language models through a new framework: The graph of thought. arXiv preprint arXiv:2308.08614. — [arXiv:2308.08614](https://arxiv.org/abs/2308.08614) find
  15. Sania Nayab, Giulio Rossolini, 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
  16. Xuefei Ning, Zinan Lin, Zixuan Zhou, Zifu Wang, Huazhong Yang, and Yu Wang. 2023. Skeleton-of-thought: Large language models can do parallel decoding. Proceedings ENLSP-III. find
  17. OpenAI. 2024. Openai o1 system card. Accessed: 2024-12-01.
  18. Philippe Remy. 2021. Name dataset. https://github.com/philipperemy/name-dataset. ↗
  19. Qwen Team. 2024. Qwq: Reflect deeply on the boundaries of the unknown. find
  20. Junlin Wang, Siddhartha Jain, Dejiao Zhang, Baishakhi Ray, Varun Kumar, and Ben Athiwaratkun. 2024. Reasoning in token economies: Budget-aware evaluation of llm reasoning strategies. arXiv preprint arXiv:2406.06461. find
  21. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2022. Self-consistency improves chain of thought reasoning in language models. arXiv preprint arXiv:2203.11171.
  22. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, brian ichter, Fei Xia, Ed 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, volume 35, pages 24824–24837. Curran Associates, Inc.
  23. An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, et al. 2024. Qwen2. 5 technical report. arXiv preprint arXiv:2412.15115. find
  24. Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Tom Griffiths, Yuan Cao, and Karthik Narasimhan. 2024. Tree of thoughts: Deliberate problem solving with large language models. Advances in Neural Information Processing Systems, 36. find
  25. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2022. React: Synergizing reasoning and acting in language models. arXiv preprint arXiv:2210.03629.
  26. Junchi Yu, Ran He, and Rex Ying. 2023. Thought propagation: An analogical approach to complex reasoning with large language models. arXiv preprint arXiv:2310.03965. — [arXiv:2310.03965](https://arxiv.org/abs/2310.03965) find
  27. 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. arXiv preprint arXiv:2309.08168.
  28. Zoom. 2025. How we’re preparing for the next era of AI. https://www.zoom.com/en/blog/ what-is-agentic-ai. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xu-2025-chain-draft-thinking-faster-writing/