---
title: "Deep Think with Confidence — paper by Fu et al. (2025)"
description: "The authors propose DeepConf, a test-time technique that uses the model's own… 99.9% accuracy. 49 references, cited by 60 papers here."
authors: ["Yichao Fu", "Xuewei Wang", "Yuandong Tian", "Jiawei Zhao"]
year: 2025
venue: "arXiv:2508.15260"
source: "https://arxiv.org/abs/2508.15260"
doi: "10.48550/arxiv.2508.15260"
arxiv: "2508.15260"
licence: "CC BY 4.0 (arXiv)"
licence_url: "http://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/fu-2025-deep-think-confidence/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/fu-2025-deep-think-confidence/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2025](https://md.research.elamur.ai/free/en/papers/2025/page/2/) / Deep Think with Confidence

# Deep Think with Confidence

**Authors:** [Yichao Fu](https://md.research.elamur.ai/free/en/authors/yichao-fu/), [Xuewei Wang](https://md.research.elamur.ai/free/en/authors/xuewei-wang/), [Yuandong Tian](https://md.research.elamur.ai/free/en/authors/yuandong-tian/) and Jiawei Zhao · **Year:** 2025 · **Venue:** arXiv:2508.15260 · **Licence:** [CC BY 4.0 (arXiv)](http://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2508.15260) · [PDF](https://arxiv.org/pdf/2508.15260) · [DOI 10.48550/arxiv.2508.15260](https://doi.org/10.48550/arxiv.2508.15260) · [arXiv 2508.15260](https://arxiv.org/abs/2508.15260) · [OpenAlex](https://openalex.org/W4416050792)

**Citations:** 1 citations (OpenAlex, 2026-10-14) · [60](https://md.research.elamur.ai/free/en/papers/fu-2025-deep-think-confidence/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
Parallel reasoning with self-consistency (sampling many chains of thought and taking a majority vote) improves LLM accuracy, but it is costly and its gains flatten as more traces are added. Part of the problem is that every trace gets an equal vote, whatever its quality. The authors propose DeepConf, a test-time technique that uses the model's own token log-probabilities to estimate how reliable each trace is. Instead of averaging confidence over a whole trace, it uses localized signals: sliding-window group confidence, the weakest-scoring segments, and confidence near the end of the trace. In offline mode, DeepConf filters and weights completed traces before voting. In online mode, it halts generation of a trace once its windowed confidence drops below a threshold calibrated on a few warmup traces. The method needs no extra training and was integrated into vLLM with small code changes. On math and science benchmarks it matches or beats plain majority voting while generating far fewer tokens. This makes heavy test-time scaling more affordable.

## Key points
- **Setup:** Benchmarks were AIME 2024/2025, HMMT 2025, BRUMO25 and GPQA-Diamond. Models were DeepSeek-8B, Qwen3-8B/32B and GPT-OSS-20B/120B. Results are averaged over 64 repetitions per setting.
- **Confidence measures:**
  - Token confidence is the negative mean log-probability of the top-k candidate tokens.
  - Traces are scored by group confidence over windows of about 1024–2048 tokens, bottom-10% group confidence, lowest group confidence, or tail confidence over the final ~2048 tokens.
  - Bottom-10% and tail confidence separate correct from incorrect traces better than the plain mean.
- **Offline result:** On AIME 2025 with GPT-OSS-120B (no tools), DeepConf@512 reached 99.9% accuracy, compared with 97.0% for cons@512 and 91.8% for pass@1.
- **Online result:** Early stopping cut generated tokens by up to 84.7% compared with full parallel thinking, with equal or better accuracy. Two variants were tested:
  - DeepConf-low keeps the top 10% of warmup traces.
  - DeepConf-high keeps the top 90%.
  - Both use 16 warmup traces, a 0.95 consensus threshold and budgets of 32–512 traces.
- **Implementation:** Only two vLLM files are modified (`logprobs.py`, `output_processor.py`). The feature is enabled per request through the OpenAI-compatible API by passing `window_size` and `threshold`.

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

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works

## Cite

Cite as: Yichao Fu, Xuewei Wang, Yuandong Tian and Jiawei Zhao. 2025. Deep Think with Confidence. arXiv:2508.15260.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@misc{fu2025deep,
  title = {{Deep Think with Confidence}},
  author = {Yichao Fu, Xuewei Wang, Yuandong Tian and Jiawei Zhao},
  year = {2025},
  doi = {10.48550/arxiv.2508.15260},
  eprint = {2508.15260},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2508.15260}
}
```

- [← previous Seed1.5-Thinking: Advancing Superb Reasoning Models with Reinforcement Learning](https://md.research.elamur.ai/free/en/papers/bytedance-seed-2025-seed1-5-thinking-advancing/)
- [next → Is Translation All You Need? A Study on Solving Multilingual Tasks with Large Language Models](https://md.research.elamur.ai/free/en/papers/liu-2025-is-translation-all-you-need/)
- [↑ back to papers of 2025](https://md.research.elamur.ai/free/en/papers/2025/page/2/)

## References

49 references, 24 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Brumo. brown university math olympiad 2025. [https://www.brumo.org/](https://www.brumo.org/), 2025. Accessed: 2025. [↗](https://www.brumo.org/)
2. Pranjal Aggarwal, Aman Madaan, Yiming Yang, et al. Let’s sample step by step: Adaptive-consistency for efficient reasoning and coding with llms. arXiv preprint arXiv:2305.11860, 2023. [find](https://scholar.google.com/scholar?q=Pranjal+Aggarwal%2C+Aman+Madaan%2C+Yiming+Yang%2C+et+al%2E+Let%E2%80%99s+sample+step+by+step%3A+Adaptive%2Dconsistency+for+efficient+reasoning+and+coding+with+llms%2E+arXiv+preprint+arXiv%3A2305%2E11860%2C+2023%2E)
3. Art of Problem Solving. 2024 aime i. [https://artofproblemsolving.com/wiki/index.php/](https://artofproblemsolving.com/wiki/index.php/) 2024_AIME_I, 2024a. Accessed: 2025. [↗](https://artofproblemsolving.com/wiki/index.php/)
4. Art of Problem Solving. 2024 aime ii. [https://artofproblemsolving.com/wiki/index.php/](https://artofproblemsolving.com/wiki/index.php/) 2024_AIME_II, 2024b. Accessed: 2025. [↗](https://artofproblemsolving.com/wiki/index.php/)
5. Art of Problem Solving. 2025 aime i. [https://artofproblemsolving.com/wiki/index.php/](https://artofproblemsolving.com/wiki/index.php/) 2025_AIME_I, 2025a. Accessed: 2025. [↗](https://artofproblemsolving.com/wiki/index.php/)
6. Art of Problem Solving. 2025 aime ii. [https://artofproblemsolving.com/wiki/index.php/](https://artofproblemsolving.com/wiki/index.php/) 2025_AIME_II, 2025b. Accessed: 2025. [↗](https://artofproblemsolving.com/wiki/index.php/)
7. Mislav Balunović, Jasper Dekoninck, Ivo Petrov, Nikola Jovanović, and Martin Vechev. [Matharena: Evaluating llms on uncontaminated math competitions](https://md.research.elamur.ai/free/en/papers/balunovic-2025-matharena-evaluating-llms-uncontaminated-math/), February 2025. URL [https://matharena.ai/](https://matharena.ai/). [↗](https://matharena.ai/)
8. Bradley Brown, Jordan Juravsky, Ryan Ehrlich, Ronald Clark, Quoc V Le, Christopher Ré, and Azalia Mirhoseini. [Large language monkeys: Scaling inference compute with repeated sampling](https://md.research.elamur.ai/free/en/papers/brown-2024-large-language-monkeys-scaling-inference/). arXiv preprint arXiv:2407.21787, 2024.
9. Lingjiao Chen, Jared Quincy Davis, Boris Hanin, Peter Bailis, Ion Stoica, Matei Zaharia, and James Zou. Are more llm calls all you need? towards scaling laws of compound inference systems, 2024a. URL https://arxiv.org/abs/2403.02419. [↗](https://arxiv.org/abs/2403.02419)
10. Xingyu Chen, Jiahao Xu, Tian Liang, Zhiwei He, Jianhui Pang, Dian Yu, Linfeng Song, Qiuzhi Liu, Mengfei Zhou, Zhuosheng Zhang, et al. Do not think that much for 2+ 3=? on the overthinking of o1-like llms. arXiv preprint arXiv:2412.[21187, 2024b](https://md.research.elamur.ai/free/en/papers/chen-2024-think-that-much-2-3/).
11. Yu-Neng Chuang, Prathusha K. Sarma, Parikshit Gopalan, John Boccio, Sara Bolouki, Xia Hu, and Helen Zhou. Learning to Route LLMs with Confidence Tokens. arXiv preprint arXiv:2410.13284, 2024. URL https://arxiv.org/abs/2410.13284. — [arXiv:2410.13284](https://arxiv.org/abs/2410.13284) [↗](https://arxiv.org/abs/2410.13284)
12. Ekaterina Fadeeva, Aleksandr Rubashevskii, Artem Shelmanov, Sergey Petrakov, Haonan Li, Hamdy Mubarak, Evgenii Tsymbalov, Gleb Kuzmin, Alexander Panchenko, Timothy Baldwin, Preslav Nakov, and Maxim Panov. [Fact-Checking the Output of Large Language Models via Token-Level Uncertainty Quantification](https://md.research.elamur.ai/free/en/papers/fadeeva-2024-fact-checking-output-large-language/). a…
13. Sebastian Farquhar, Jannik Kossen, Lorenz Kuhn, and Yarin Gal. Detecting hallucinations in large language models using semantic entropy. Nature, 630(8017):625–630, 2024. [find](https://scholar.google.com/scholar?q=Sebastian+Farquhar%2C+Jannik+Kossen%2C+Lorenz+Kuhn%2C+and+Yarin+Gal%2E+Detecting+hallucinations+in+large+language+models+using+semantic+entropy%2E+Nature%2C+630%288017%29%3A625%E2%80%93630%2C+2024%2E)
14. Yichao Fu, Junda Chen, Siqi Zhu, Zheyu Fu, Zhongdongming Dai, Yonghao Zhuang, Yian Ma, Aurick Qiao, Tajana Rosing, Ion Stoica, et al. Efficiently scaling llm reasoning with certaindex. arXiv preprint arXiv:2412.20993, 2024. — [arXiv:2412.20993](https://arxiv.org/abs/2412.20993) [find](https://scholar.google.com/scholar?q=Yichao+Fu%2C+Junda+Chen%2C+Siqi+Zhu%2C+Zheyu+Fu%2C+Zhongdongming+Dai%2C+Yonghao+Zhuang%2C+Yian+Ma%2C+Aurick+Qiao%2C+Tajana+Rosing%2C+Ion+Stoica%2C+et+al%2E+Efficiently+scaling+llm+reasoning+with+certaindex%2E+arXiv+preprint+)
15. Yichao Fu, Junda Chen, Yonghao Zhuang, Zheyu Fu, Ion Stoica, and Hao Zhang. Reasoning without self-doubt: More efficient chain-of-thought through certainty probing. In ICLR 2025 Workshop on Foundation Models in the Wild, 2025. [find](https://scholar.google.com/scholar?q=Yichao+Fu%2C+Junda+Chen%2C+Yonghao+Zhuang%2C+Zheyu+Fu%2C+Ion+Stoica%2C+and+Hao+Zhang%2E+Reasoning+without+self%2Ddoubt%3A+More+efficient+chain%2Dof%2Dthought+through+certainty+probing%2E+In+ICLR+2025+Workshop+on+Foundation)
16. Jiahui Geng, Fengyu Cai, Yuxia Wang, Heinz Koeppl, Preslav Nakov, and Iryna Gurevych. [A survey of confidence estimation and calibration in large language models](https://md.research.elamur.ai/free/en/papers/geng-2023-survey-confidence-estimation-calibration-large/). In Juan Li, Xiang Li, and Haizhou Li (eds.), Proceedings of the 2024 Conference of the North American Chapter of the Association for Compu…
17. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. [Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning](https://md.research.elamur.ai/free/en/papers/deepseek-ai-2025-deepseek-r1-incentivizing-reasoning-capability/). arXiv preprint arXiv:2501.12948, 2025.
18. Zhijian Han, Zhi Li, Yutong Wang, Chenglu Guo, Ruifei Song, Jiacheng He, Jun Pan, Jianing Wu, Shuai Li, Liang Gao, and Weihang Chen. [Adaptive Inference-Time Compute: LLMs Can Predict if They Can Do Better, Even Mid-Generation](https://md.research.elamur.ai/free/en/papers/manvi-2024-adaptive-inference-time-compute-llms/). arXiv preprint arXiv:2410.02725, 2024. URL https://arxiv.org/abs/2410.027… [↗](https://arxiv.org/abs/2410.027…)
19. Michael Hassid, Gabriel Synnaeve, Yossi Adi, and Roy Schwartz. Don’t overthink it. preferring shorter thinking chains for improved llm reasoning. arXiv preprint arXiv:2505.17813, 2025. — [arXiv:2505.17813](https://arxiv.org/abs/2505.17813) [find](https://scholar.google.com/scholar?q=Michael+Hassid%2C+Gabriel+Synnaeve%2C+Yossi+Adi%2C+and+Roy+Schwartz%2E+Don%E2%80%99t+overthink+it%2E+preferring+shorter+thinking+chains+for+improved+llm+reasoning%2E+arXiv+preprint+arXiv%3A2505%2E17813%2C+2025%2E+%E2%80%94+%5BarXiv%3A2505%2E1)
20. HMMT. Hmmt 2025. [https://www.hmmt.org/](https://www.hmmt.org/), 2025. Accessed: 2025. [↗](https://www.hmmt.org/)
21. Bairu Hou, Yang Zhang, Jiabao Ji, Yujian Liu, Kaizhi Qian, Jacob Andreas, and Shiyu Chang. [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/). arXiv preprint arXiv:2504.01296, 2025.
22. Robert Irvine, Douglas Boubert, Vyas Raina, Adian Liusie, Ziyi Zhu, Vineet Mudupalli, Aliaksei Korshuk, Zongyi Liu, Fritz Cremer, Valentin Assassi, et al. [Rewarding chatbots for real-world engagement with millions of users](https://md.research.elamur.ai/free/en/papers/irvine-2023-rewarding-chatbots-real-world-engagement/). arXiv preprint arXiv:2303.06135, 2023.
23. Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, Aleksander Madry, Alex Beutel, Alex Carney, et al. [Openai o1 system card](https://md.research.elamur.ai/free/en/papers/openai-2024-openai-o1-system-card/). arXiv preprint arXiv:2412.16720, 2024.
24. Siddhartha Jain, Xiaofei Ma, Anoop Deoras, and Bing Xiang. Lightweight reranking for language model generations. In Lun-Wei Ku, Andre Martins, and Vivek Srikumar (eds.), Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 6960–6984, Ba… [find](https://scholar.google.com/scholar?q=Siddhartha+Jain%2C+Xiaofei+Ma%2C+Anoop+Deoras%2C+and+Bing+Xiang%2E+Lightweight+reranking+for+language+model+generations%2E+In+Lun%2DWei+Ku%2C+Andre+Martins%2C+and+Vivek+Srikumar+%28eds%2E%29%2C+Proceedings+of+the+62nd+Annual)
25. Zhewei Kang, Xuandong Zhao, and Dawn Song. Scalable best-of-n selection for large language models via self-certainty, 2025. URL http://arxiv.org/abs/2502.18581. [↗](http://arxiv.org/abs/2502.18581)
26. Yiwei Li, Peiwen Yuan, Shaoxiong Feng, Boyuan Pan, Xinglin Wang, Bin Sun, Heda Wang, and Kan Li. Escape sky-high cost: Early-stopping self-consistency for multi-step reasoning. arXiv preprint arXiv:2401.10480, 2024. [find](https://scholar.google.com/scholar?q=Yiwei+Li%2C+Peiwen+Yuan%2C+Shaoxiong+Feng%2C+Boyuan+Pan%2C+Xinglin+Wang%2C+Bin+Sun%2C+Heda+Wang%2C+and+Kan+Li%2E+Escape+sky%2Dhigh+cost%3A+Early%2Dstopping+self%2Dconsistency+for+multi%2Dstep+reasoning%2E+arXiv+preprint+arXiv%3A24)
27. Haotian Luo, Li Shen, Haiying He, Yibo Wang, Shiwei Liu, Wei Li, Naiqiang Tan, Xiaochun Cao, and Dacheng Tao. O1-pruner: Length-harmonizing fine-tuning for o1-like reasoning pruning. arXiv preprint arXiv:2501.12570, 2025. [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+O1%2Dpruner%3A+Length%2Dharmonizing+fine%2Dtuning+for+o1%2Dlike+reasoning+pruning%2E+arXiv+preprint+ar)
28. Yingqian Min, Zhipeng Chen, Jinhao Jiang, Jie Chen, Jia Deng, Yiwen Hu, Yiru Tang, Jiapeng Wang, Xiaoxue Cheng, Huatong Song, et al. [Imitate, explore, and self-improve: A reproduction report on slow-thinking reasoning systems](https://md.research.elamur.ai/free/en/papers/min-2024-imitate-explore-self-improve-reproduction/). arXiv preprint arXiv:2412.09413, 2024.
29. Niklas Muennighoff, Zitong Yang, Weijia Shi, Xiang Lisa Li, Li Fei-Fei, Hannaneh Hajishirzi, Luke Zettlemoyer, Percy Liang, Emmanuel Candès, and Tatsunori Hashimoto. [s1: Simple test-time scaling](https://md.research.elamur.ai/free/en/papers/muennighoff-2025-s1-simple-test-time-scaling/). arXiv preprint arXiv:2501.19393, 2025.
30. OpenAI. Introducing gpt-5, 2025. URL [https://openai.com/index/introducing-gpt-5/](https://openai.com/index/introducing-gpt-5/). Accessed: 2025-08-09. [↗](https://openai.com/index/introducing-gpt-5/)
31. OpenAI. [gpt-oss-120b & gpt-oss-20b model card](https://md.research.elamur.ai/free/en/papers/agarwal-2025-gpt-oss-120b-gpt-oss/). Model card, OpenAI, August 2025. URL [https://cdn.openai.com/pdf/419b6906-9da6-406c-a19d-1bb078ac7637/](https://cdn.openai.com/pdf/419b6906-9da6-406c-a19d-1bb078ac7637/) oai_gpt-oss_model_card.pdf. [↗](https://cdn.openai.com/pdf/419b6906-9da6-406c-a19d-1bb078ac7637/)
32. Soumyasundar Pal, Didier Chételat, Yingxue Zhang, and Mark Coates. Hint Marginalization for Improved Reasoning in Large Language Models. OpenReview, 2024. URL [https://openreview.net/forum?id=DzKdjWe59v](https://openreview.net/forum?id=DzKdjWe59v). [↗](https://openreview.net/forum?id=DzKdjWe59v)
33. David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R Bowman. [Gpqa: A graduate-level google-proof q&a benchmark](https://md.research.elamur.ai/free/en/papers/rein-2023-gpqa-graduate-level-google-proof/). In First Conference on Language Modeling, 2024.
34. Jie Ren, Yao Zhao, Tu Vu, Peter J. Liu, and Balaji Lakshminarayanan. Self-Evaluation Improves Selective Generation in Large Language Models. arXiv preprint arXiv:2312.09300, dec 2023. URL https://arxiv.org/abs/2312.09300. — [arXiv:2312.09300](https://arxiv.org/abs/2312.09300) [↗](https://arxiv.org/abs/2312.09300)
35. Charlie Snell, Jaehoon Lee, Kelvin Xu, and Aviral Kumar. [Scaling llm test-time compute optimally can be more effective than scaling model parameters](https://md.research.elamur.ai/free/en/papers/snell-2024-scaling-llm-test-time-compute/). arXiv preprint arXiv:2408.03314, 2024.
36. Kimi Team, Angang Du, Bofei Gao, Bowei Xing, Changjiu Jiang, Cheng Chen, Cheng Li, Chenjun Xiao, Chenzhuang Du, Chonghua Liao, et al. Kimi k1. 5: Scaling reinforcement learning with llms. arXiv preprint arXiv:2501.[12599, 2025](https://md.research.elamur.ai/free/en/papers/kimi-team-2025-kimi-k1-5-scaling-reinforcement/).
37. Vernon Y.H. Toh, Deepanway Ghosal, and Soujanya Poria. Not All Votes Count! Programs as Verifiers Improve Self-Consistency of Language Models for Math Reasoning. arXiv preprint arXiv:2410.12608, 2024. URL https://arxiv.org/abs/2410.12608. [↗](https://arxiv.org/abs/2410.12608)
38. Guangya Wan, Yuqi Wu, Jie Chen, and Sheng Li. [Reasoning aware self-consistency: Leveraging reasoning paths for efficient LLM sampling](https://md.research.elamur.ai/free/en/papers/wan-2025-reasoning-aware-self-consistency-leveraging/). In Luis Chiruzzo, Alan Ritter, and Lu Wang (eds.), Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Lin…
39. Han Wang, Archiki Prasad, Elias Stengel-Eskin, and Mohit Bansal. Soft Self-Consistency Improves Language Model Agents. arXiv preprint arXiv:2402.13212, 2024. URL [https://arxiv.org/](https://arxiv.org/) abs/2402.13212. [↗](https://arxiv.org/)
40. Weiqin Wang, Yile Wang, and Hui Huang. Ranked voting based self-consistency of large language models. In Wanxiang Che, Joyce Nabende, Ekaterina Shutova, and Mohammad Taher Pilehvar (eds.), Findings of the Association for Computational Linguistics: ACL 2025, pp. 14410–14426, Vienna, Austria, July 202… [find](https://scholar.google.com/scholar?q=Weiqin+Wang%2C+Yile+Wang%2C+and+Hui+Huang%2E+Ranked+voting+based+self%2Dconsistency+of+large+language+models%2E+In+Wanxiang+Che%2C+Joyce+Nabende%2C+Ekaterina+Shutova%2C+and+Mohammad+Taher+Pilehvar+%28eds%2E%29%2C+Findings+of)
41. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed H Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. [Self-Consistency Improves Chain of Thought Reasoning in Language Models](https://md.research.elamur.ai/free/en/papers/wang-2022-self-consistency-improves-chain-thought/). arXiv preprint arXiv:2203.11171, 2023. URL https://arxiv.org/abs/2203. 11171. [↗](https://arxiv.org/abs/2203)
42. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. [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/). Advances in neural information processing systems, 35:24824–24837, 2022.
43. Sean Welleck, Amanda Bertsch, Matthew Finlayson, Hailey Schoelkopf, Alex Xie, Graham Neubig, Ilia Kulikov, and Zaid Harchaoui. [From decoding to meta-generation: Inference-time algorithms for large language models](https://md.research.elamur.ai/free/en/papers/welleck-2024-decoding-meta-generation-inference-time/). arXiv preprint arXiv:2406.16838, 2024.
44. Yangzhen Wu, Zhiqing Sun, Shanda Li, Sean Welleck, and Yiming Yang. [Inference scaling laws: An empirical analysis of compute-optimal inference for problem-solving with language models](https://md.research.elamur.ai/free/en/papers/wu-2024-inference-scaling-laws-empirical-analysis/). arXiv preprint arXiv:2408.00724, 2024. xAI. Grok 4, 2025. URL [https://x.ai/news/grok-4](https://x.ai/news/grok-4). Accessed: 2025-08-10. [↗](https://x.ai/news/grok-4)
45. Mingfeng Xue, Dayiheng Liu, Wenqiang Lei, Xingzhang Ren, Baosong Yang, Jun Xie, Yidan Zhang, Dezhong Peng, and Jiancheng Lv. Dynamic Voting for Efficient Reasoning in Large Language Models. In Findings of the Association for Computational Linguistics: EMNLP 2023, pp. 3085–3104. Association for Compu… [find](https://scholar.google.com/scholar?q=Mingfeng+Xue%2C+Dayiheng+Liu%2C+Wenqiang+Lei%2C+Xingzhang+Ren%2C+Baosong+Yang%2C+Jun+Xie%2C+Yidan+Zhang%2C+Dezhong+Peng%2C+and+Jiancheng+Lv%2E+Dynamic+Voting+for+Efficient+Reasoning+in+Large+Language+Models%2E+In+Finding)
46. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. [Qwen3 technical report](https://md.research.elamur.ai/free/en/papers/yang-2025-qwen3-technical-report/). arXiv preprint arXiv:2505.09388, 2025a.
47. Chenxu Yang, Qingyi Si, Yongjie Duan, Zheliang Zhu, Chenyu Zhu, Qiaowei Li, Zheng Lin, Li Cao, and Weiping Wang. [Dynamic early exit in reasoning models](https://md.research.elamur.ai/free/en/papers/yang-2025-dynamic-early-exit-reasoning-models/). arXiv preprint arXiv:2504.15895, 2025b.
48. Yixin Ye, Zhen Huang, Yang Xiao, Ethan Chern, Shijie Xia, and Pengfei Liu. [Limo: Less is more for reasoning](https://md.research.elamur.ai/free/en/papers/ye-2025-limo-less-more-reasoning/). arXiv preprint arXiv:2502.03387, 2025.
49. Xuandong Zhao, Zhewei Kang, Aosong Feng, Sergey Levine, and Dawn Song. Learning to Reason without External Rewards. arXiv preprint arXiv:2505.19590, 2025. URL [https://arxiv.org/](https://arxiv.org/) abs/2505.19590. [↗](https://arxiv.org/)

*Updated 2 November 2026*