Elamur Research⌕
☰
home / papers / 2026 / GrACE: A Generative Approach to Better Confidence Elicitation and Efficient Test-Time Scaling in Large Language Models

GrACE: A Generative Approach to Better Confidence Elicitation and Efficient Test-Time Scaling in Large Language Models

Paper · 2026 · Zhaohan Zhang, Ziquan Liu and Ioannis Patras

Summary

Eliciting confidence from LLMs is costly when it requires a second evaluation pass or an auxiliary model, and verbalized confidence tends to be vague and overconfident. GrACE (Generative cAlibrated Confidence Elicitation) avoids both problems. It extends the model's vocabulary with a dedicated <CNF> token and trains the model to emit it at the end of each answer. The confidence score is the similarity between the final hidden state and the <CNF> embedding, so it comes out during generation at almost no extra cost.

Calibration targets are built by probing the model's internal confidence, binning training samples with k-fold grouping, and assigning each sample its group's accuracy. Only LoRA weights and the <CNF> embedding are fine-tuned. On open-ended generation, GrACE gives better-calibrated and more discriminative confidence than six baselines and generalizes to unseen domains. The authors also use its scores for test-time scaling, through confidence-weighted voting and early stopping. This raises accuracy while cutting the number of samples needed, which makes reliable real-time confidence practical to deploy.

Key points

  • Compared with the strongest baseline, Apricot, GrACE improves ECE by 2.6%, Brier score by 1.8%, and AUROC by 4.4%.
  • It needs no additional sampling, no auxiliary model, and no post-generation evaluation stage.
  • Two test-time scaling strategies are proposed:
    • GrACE-SC: self-consistency in which each sample's vote is weighted by its confidence.
    • GrACE-ES: early stopping once a sample's confidence passes a threshold.
    • Together they raise accuracy by 1.54% on average.
  • At a sampling budget of 8, GrACE-SC beats plain self-consistency on MathQA and ARC_C:
    • by 2.6 and 0.9 points with Llama3.1-8B-Instruct;
    • by 3.3 and 0.8 points with Qwen2.5-7B-Instruct.
  • With GrACE-ES, 38.6% of MathQA queries and 80.1% of ARC_C queries stop after a single sample.
  • The main limitation is that GrACE gives one overall score per response, not a score for each claim or reasoning step.

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

🎓Queen Mary University of LondonLondon · United Kingdom57 works · 3,600 citations

 Fields, topics and concepts

📚 Shelves 4The 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 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,894 works

Cite

Cite as Zhaohan Zhang, Ziquan Liu and Ioannis Patras. 2026. GrACE: A Generative Approach to Better Confidence Elicitation and Efficient Test-Time Scaling in Large Language Models. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.1069

@inproceedings{zhang2026grace,
  title = {{GrACE: A Generative Approach to Better Confidence Elicitation and Efficient Test-Time Scaling in Large Language Models}},
  author = {Zhaohan Zhang, Ziquan Liu and Ioannis Patras},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1069},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1069}
}

References

56 references, 29 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Marah Abdin, Jyoti Aneja, Hany Awadalla, Ahmed Awadallah, Ammar Ahmad Awan, Nguyen Bach, Amit Bahree, Arash Bakhtiari, Jianmin Bao, Harkirat Behl, and 1 others. 2024. Phi-3 technical report: A highly capable language model locally on your phone. arXiv preprint arXiv:2404.14219.
  2. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, and 1 others. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774.
  3. Pranjal Aggarwal, Aman Madaan, Yiming Yang, and 1 others. Let’s sample step by step: Adaptive-consistency for efficient reasoning and coding with llms. In The 2023 Conference on Empirical Methods in Natural Language Processing. find
  4. Aida Amini, Saadia Gabriel, Shanchuan Lin, Rik Koncel-Kedziorski, Yejin Choi, and Hannaneh Hajishirzi. 2019. Mathqa: Towards interpretable math word problem solving with operation-based formalisms. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computation…
  5. Anthropic. 2024. Claude 3: A conversational ai model. find
  6. Amos Azaria and Tom Mitchell. The internal state of an llm knows when it’s lying. In The 2023 Conference on Empirical Methods in Natural Language Processing.
  7. Neil Band, Xuechen Li, Tengyu Ma, and Tatsunori Hashimoto. 2024. Linguistic calibration of long-form generations. In Proceedings of the 41st International Conference on Machine Learning, pages 2732–2778. find
  8. Glenn W Brier. 1950. Verification of forecasts expressed in terms of probability. Monthly weather review, 78(1):1–3. find
  9. Bradley Brown, Jordan Juravsky, Ryan Ehrlich, Ronald Clark, Quoc V Le, Christopher Ré, and Azalia Mirhoseini. 2024. Large language monkeys: Scaling inference compute with repeated sampling. arXiv preprint arXiv:2407.21787.
  10. Xingyu Chen, Jiahao Xu, Tian Liang, Zhiwei He, Jianhui Pang, Dian Yu, Linfeng Song, Qiuzhi Liu, Mengfei Zhou, Zhuosheng Zhang, and 1 others. 2024. Do not think that much for 2+ 3=? on the overthinking of o1-like llms. arXiv preprint arXiv:2412.21187.
  11. Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. 2018. Think you have solved question answering? try arc, the ai2 reasoning challenge. arXiv preprint arXiv:1803.05457.
  12. Roi Cohen, Konstantin Dobler, Eden Biran, and Gerard de Melo. 2024. I don’t know: Explicit modeling of uncertainty with an [idk] token. Advances in Neural Information Processing Systems, 37:10935–10958. find
  13. Shrey Desai and Greg Durrett. 2020. Calibration of pre-trained transformers. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 295–302. find
  14. Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, and 1 others. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783.
  15. Yichao Fu, Xuewei Wang, Yuandong Tian, and Jiawei Zhao. 2025. Deep think with confidence. arXiv preprint arXiv:2508.15260.
  16. Sachin Goyal, Ziwei Ji, Ankit Singh Rawat, Aditya Krishna Menon, Sanjiv Kumar, and Vaishnavh Nagarajan. Think before you speak: Training language models with pause tokens. In The Twelfth International Conference on Learning Representations. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav… — The Llama 3 Herd of Models
  17. Lorenz Kuhn, Yarin Gal, and Sebastian Farquhar. Semantic uncertainty: Linguistic invariances for uncertainty estimation in natural language generation. In The Eleventh International Conference on Learning Representations.
  18. Chuan Guo, Geoff Pleiss, Yu Sun, and Kilian Q Weinberger. 2017. On calibration of modern neural networks. In International conference on machine learning, pages 1321–1330. PMLR. find
  19. Volodymyr Kuleshov and Percy S Liang. 2015. Calibrated structured prediction. Advances in Neural Information Processing Systems, 28. find
  20. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, and 1 others. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948.
  21. Chengzu Li, Han Zhou, Goran Glavaš, Anna Korhonen, and Ivan Vulić. 2025. Large language models are miscalibrated in-context learners. In Findings of the Association for Computational Linguistics: ACL 2025, pages 11575–11596. find
  22. Sophia Hager, David Mueller, Kevin Duh, and Nicholas Andrews. 2025. Uncertainty distillation: Teaching language models to express semantic confidence. arXiv preprint arXiv:2503.14749.
  23. Edward J Hu, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, and 1 others. Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations.
  24. Chengsong Huang, Langlin Huang, Jixuan Leng, Jiacheng Liu, and Jiaxin Huang. 2025. Efficient test-time scaling via self-calibration. arXiv preprint arXiv:2503.00031. — [arXiv:2503.00031](https://arxiv.org/abs/2503.00031) find
  25. Ziwei Ji, Lei Yu, Yeskendir Koishekenov, Yejin Bang, Anthony Hartshorn, Alan Schelten, Cheng Zhang, Pascale Fung, and Nicola Cancedda. 2025. Calibrating verbal uncertainty as a linear feature to reduce hallucinations. arXiv preprint arXiv:2503.14477.
  26. Mandar Joshi, Eunsol Choi, Daniel S Weld, and Luke Zettlemoyer. 2017. Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1601–1611.
  27. Saurav Kadavath, Tom Conerly, Amanda Askell, Tom Henighan, Dawn Drain, Ethan Perez, Nicholas Schiefer, Zac Hatfield-Dodds, Nova DasSarma, Eli Tran-Johnson, and 1 others. 2022. Language models (mostly) know what they know. arXiv preprint arXiv:2207.05221. find
  28. Sanyam Kapoor, Nate Gruver, Manley Roberts, Katherine M Collins, Arka Pal, Umang Bhatt, Adrian Weller, Samuel Dooley, Micah Goldblum, and Andrew Gordon Wilson. Large language models must be taught to know what they don’t know. In The Thirty-eighth Annual Conference on Neural Information Processing S… find
  29. Jannik Kossen, Jiatong Han, Muhammed Razzak, Lisa Schut, Shreshth Malik, and Yarin Gal. 2024. Semantic entropy probes: Robust and cheap hallucination detection in llms. arXiv preprint arXiv:2406.15927. — [arXiv:2406.15927](https://arxiv.org/abs/2406.15927) find
  30. 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. In The Twelfth International Conference on Learning Representations. find
  31. Chin-Yew Lin. 2004. Rouge: A package for automatic evaluation of summaries. In Text summarization branches out, pages 74–81.
  32. Xin Liu, Farima Fatahi Bayat, and Lu Wang. 2024. Enhancing language model factuality via activation-based confidence calibration and guided decoding. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 10436–10448. find
  33. Varun Magesh, Faiz Surani, Matthew Dahl, Mirac Suzgun, Christopher D Manning, and Daniel E Ho. 2024. Hallucination-free? assessing the reliability of leading ai legal research tools. arXiv preprint arXiv:2405.20362.
  34. Andrey Malinin and Mark Gales. Uncertainty estimation in autoregressive structured prediction. In International Conference on Learning Representations. find
  35. Niklas Muennighoff, Zitong Yang, Weijia Shi, Xiang Lisa Li, Li Fei-Fei, Hannaneh Hajishirzi, Luke Zettlemoyer, Percy Liang, Emmanuel Candès, and Tatsunori Hashimoto. 2025. s1: Simple test-time scaling. arXiv preprint arXiv:2501.19393.
  36. Mahdi Pakdaman Naeini, Gregory Cooper, and Milos Hauskrecht. 2015. Obtaining well calibrated probabilities using bayesian binning. In Proceedings of the AAAI conference on artificial intelligence, volume 29. find
  37. Alexandru Niculescu-Mizil and Rich Caruana. 2005. Predicting good probabilities with supervised learning. In Proceedings of the 22nd international conference on Machine learning, pages 625–632. find
  38. John Platt and 1 others. 1999. Probabilistic outputs for support vector machines and comparisons to regularized likelihood methods. Advances in large margin classifiers, 10(3):61–74. find
  39. Karan Singhal, Shekoofeh Azizi, Tao Tu, S Sara Mahdavi, Jason Wei, Hyung Won Chung, Nathan Scales, Ajay Tanwani, Heather Cole-Lewis, Stephen Pfohl, and 1 others. 2023. Large language models encode clinical knowledge. Nature, 620(7972):172–180.
  40. Charlie Snell, Jaehoon Lee, Kelvin Xu, and Aviral Kumar. 2024. Scaling llm test-time compute optimally can be more effective than scaling model parameters. arXiv preprint arXiv:2408.03314.
  41. Hexiang Tan, Fei Sun, Sha Liu, Du Su, Qi Cao, Xin Chen, Jingang Wang, Xunliang Cai, Yuanzhuo Wang, Huawei Shen, and 1 others. 2025. Too consistent to detect: A study of self-consistent errors in llms. arXiv preprint arXiv:2505.17656.
  42. Amir Taubenfeld, Tom Sheffer, Eran Ofek, Amir Feder, Ariel Goldstein, Zorik Gekhman, and Gal Yona. 2025. Confidence improves self-consistency in llms. arXiv preprint arXiv:2502.06233. find
  43. Katherine Tian, Eric Mitchell, Allan Zhou, Archit Sharma, Rafael Rafailov, Huaxiu Yao, Chelsea Finn, and Christopher D Manning. 2023. Just ask for calibration: Strategies for eliciting calibrated confidence scores from language models fine-tuned with human feedback. In Proceedings of the 2023 Confer… find
  44. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, and 1 others. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288.
  45. Dennis Thomas Ulmer, Martin Gubri, Hwaran Lee, Sangdoo Yun, and Seong Joon Oh. 2024. Calibrating large language models using their generations only. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics, pages 15440–15459. Association for Computational Linguistic… find
  46. Qianqian Xie, Weiguang Han, Zhengyu Chen, Ruoyu Xiang, Xiao Zhang, Yueru He, Mengxi Xiao, Dong Li, Yongfu Dai, Duanyu Feng, and 1 others. 2024. Finben: A holistic financial benchmark for large language models. Advances in Neural Information Processing Systems, 37:95716–95743. find
  47. Miao Xiong, Zhiyuan Hu, Xinyang Lu, YIFEI LI, Jie Fu, Junxian He, and Bryan Hooi. Can llms express their uncertainty? an empirical evaluation of confidence elicitation in llms. In The Twelfth International Conference on Learning Representations. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyu… — Qwen3 Technical Report
  48. An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, and 1 others. 2024a. Qwen2. 5 technical report. arXiv preprint arXiv:2412.15115. find
  49. Ruihan Yang, Caiqi Zhang, Zhisong Zhang, Xinting Huang, Sen Yang, Nigel Collier, Dong Yu, and Deqing Yang. 2024b. Logu: Long-form generation with uncertainty expressions. arXiv preprint arXiv:2410.14309.
  50. Ruihan Yang, Caiqi Zhang, Zhisong Zhang, Xinting Huang, Dong Yu, Nigel Collier, and Deqing Yang. 2025b. Uncle: Uncertainty expressions in long-form generation. arXiv preprint arXiv:2505.16922. — [arXiv:2505.16922](https://arxiv.org/abs/2505.16922) find
  51. Caiqi Zhang, Ruihan Yang, Zhisong Zhang, Xinting Huang, Sen Yang, Dong Yu, and Nigel Collier. 2024. Atomic calibration of llms in long-form generations. arXiv preprint arXiv:2410.13246. find
  52. Zhaohan Zhang, Ziquan Liu, and Ioannis Patras. 2025. Get confused cautiously: Textual sequence memorization erasure with selective entropy maximization. In Proceedings of the 31st International Conference on Computational Linguistics, pages 10924–10939.
  53. Guangya Wan, Yuqi Wu, Jie Chen, and Sheng Li. 2025. Reasoning aware self-consistency: Leveraging reasoning paths for efficient llm sampling. In The 2025 Annual Conference of the Nations of the Americas Chapter of the ACL.
  54. Xinyi Wang, Lucas Caccia, Oleksiy Ostapenko, Xingdi Yuan, William Yang Wang, and Alessandro Sordoni. Guiding language model reasoning with planning tokens. In First Conference on Language Modeling. find
  55. 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.
  56. Johannes Welbl, Nelson F Liu, and Matt Gardner. 2017. Crowdsourcing multiple choice science questions. In Proceedings of the 3rd Workshop on Noisy User-generated Text, pages 94–106.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2026-grace-generative-approach-better-confidence/