Elamur Research⌕
☰
home / papers / 2025 / Seeing is Believing, but How Much? A Comprehensive Analysis of Verbalized Calibration in Vision-Language Models

Seeing is Believing, but How Much? A Comprehensive Analysis of Verbalized Calibration in Vision-Language Models

Paper · 2025 · Weihao Xuan, Qingcheng Zeng, Heli Qi et al.

Summary

Vision-language models (VLMs) can state in plain language how sure they are of an answer. This study asks whether that stated confidence actually tracks how often they are correct. The authors evaluate verbalized calibration across three model families: instruction-tuned, text-reasoning, and visual-reasoning ("thinking with images") systems, both commercial and open-source. The tests cover four task areas: image reasoning, video reasoning, visual factuality, and visual math. Each is run under three input scenarios: text prompts accompanying images, questions rendered inside the image itself, and identical content shown either as text or as a picture. Calibration is measured mainly with Expected Calibration Error (ECE). Most VLMs are clearly miscalibrated. OpenAI's image-reasoning models o3 and o4-mini align confidence with accuracy much better across benchmarks and modalities, which suggests that reasoning grounded in the visual input helps. The authors also propose Visual Confidence-Aware Prompting (VCAP). It first asks the model how confident it is in its visual interpretation, then asks for the final answer. VCAP outperforms Top-K and self-reflection prompting. These findings matter for deploying multimodal assistants safely.

Key points

  • Benchmarks used:
    • MMMU-Pro, for image reasoning and, in its vision-only variant, for the embedded-instruction setting.
    • VideoMMMU, with 32 uniformly sampled frames per video.
    • Visual SimpleQA, for factuality.
    • MathVista and MathVision, testmini splits, for math.
    • IsoBench, for semantically aligned text-vs-image inputs across mathematics, games, science, and algorithms.
  • Evaluated models include Qwen2.5-VL (7B, 72B), InternVL3 78B, Skywork-R1V2 38B, o3, and o4-mini.
  • Miscalibration appears consistently across tasks and input formats. The "think with images" models are the notable exception.
  • VCAP is compared against Top-K (K=3) and self-reflection prompting and gives the best calibration for most models. For the already well-calibrated o4-mini, it still improves on vanilla prompting, though the gain is smaller.
  • Limitations noted by the authors:
    • The best-calibrated models are closed-source, so the reasons for their advantage cannot be analyzed.
    • Only two production-level image-reasoning models exist, which prevents comparison across model families.
    • Scenarios such as temporal visual grounding in video were not studied.

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

🎓The University of TokyoTokyo · Japan90 works · 9,156 citations🎓Northwestern UniversityEvanston · United States41 works · 2,194 citations🔬RIKEN Center for Advanced Intelligence ProjectTokyo · Japan15 works · 957 citations🎓Waseda UniversityTokyo · Japan20 works · 584 citations🔬RIKENWako · Japan26 works · 526 citations

 Fields, topics and concepts

📚 Shelves 3The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 3Research subjects and languages the work is about.Language models & pretraining · 3,320 worksReasoning · 1,426 worksVision & language · 657 works

Cite

Cite as Weihao Xuan, Qingcheng Zeng, Heli Qi, Junjue Wang and Naoto Yokoya. 2025. Seeing is Believing, but How Much? A Comprehensive Analysis of Verbalized Calibration in Vision-Language Models. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.emnlp-main.74

@inproceedings{xuan2025seeing,
  title = {{Seeing is Believing, but How Much? A Comprehensive Analysis of Verbalized Calibration in Vision-Language Models}},
  author = {Weihao Xuan, Qingcheng Zeng, Heli Qi, Junjue Wang and Naoto Yokoya},
  booktitle = {Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing},
  year = {2025},
  doi = {10.18653/v1/2025.emnlp-main.74},
  url = {https://doi.org/10.18653/v1/2025.emnlp-main.74}
}

References

44 references, 17 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Pravesh Agrawal, Szymon Antoniak, Emma Bou Hanna, Baptiste Bout, Devendra Chaplot, Jessica Chudnovsky, Diogo Costa, Baudouin De Monicault, Saurabh Garg, Theophile Gervet, Soham Ghosh, Amélie Héliou, Paul Jacob, Albert Q. Jiang, Kartik Khandelwal, Timothée Lacroix, Guillaume Lample, Diego Las Casas,… find
  2. Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, and 8 others. 2025. Qwen2.5-vl technical report. Preprint, arXiv:2502.13…
  3. Umesh Bodhwani, Yuan Ling, Shujing Dong, Yarong Feng, Hongfei Li, and Ayush Goyal. 2025. A calibrated reflection approach for enhancing confidence estimation in LLMs. In Proceedings of the 5th Workshop on Trustworthy NLP (TrustNLP 2025), pages 399–411, Albuquerque, New Mexico. Association for Comput… find
  4. Mirko Borszukovszki, Ivo Pascal De Jong, and Matias Valdenegro-Toro. 2025. Know what you do not know: Verbalized uncertainty estimation robustness on corrupted images in vision-language models. In Proceedings of the 5th Workshop on Trustworthy NLP (TrustNLP 2025), pages 247–265, Albuquerque, New Mex… find
  5. Yiming Chen, Xianghu Yue, Chen Zhang, Xiaoxue Gao, Robby T. Tan, and Haizhou Li. 2024. Voicebench: Benchmarking llm-based voice assistants. Preprint, arXiv:2410.17196.
  6. Zijun Chen, Wenbo Hu, Guande He, Zhijie Deng, ZHeng ZHang, and Richang Hong. 2025. Unveiling uncertainty: A deep dive into calibration and performance of multimodal large language models. In Proceedings of the 31st International Conference on Computational Linguistics, pages 3095–3109, Abu Dhabi, UA… find
  7. 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. 2024. Fact-checking the output of large language models via token-level uncertainty quantificat…
  8. Ekaterina Fadeeva, Roman Vashurin, Akim Tsvigun, Artem Vazhentsev, Sergey Petrakov, Kirill Fedyanin, Daniil Vasilev, Elizaveta Goncharova, Alexander Panchenko, Maxim Panov, Timothy Baldwin, and Artem Shelmanov. 2023. LM-polygraph: Uncertainty estimation for language models. In Proceedings of the 202… find
  9. Sebastian Farquhar, Jannik Kossen, Lorenz Kuhn, and Yarin Gal. 2024. Detecting hallucinations in large language models using semantic entropy. Nature, 630(8017):625–630. find
  10. Deqing Fu, Ruohao Guo, Ghazal Khalighinejad, Ollie Liu, Bhuwan Dhingra, Dani Yogatama, Robin Jia, and Willie Neiswanger. 2024. Isobench: Benchmarking multimodal foundation models on isomorphic representations. Preprint, arXiv:2404.01266. — [arXiv:2404.01266](https://arxiv.org/abs/2404.01266) find
  11. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Arthur Hinsvark, and 542 o… — The Llama 3 Herd of Models
  12. Tobias Groot and Matias Valdenegro Toro. 2024. Overconfidence is key: Verbalized uncertainty evaluation in large language and vision-language models. In Proceedings of the 4th Workshop on Trustworthy Natural Language Processing (TrustNLP 2024), pages 145–171, Mexico City, Mexico. Association for Com… find
  13. Sophia Hager, David Mueller, Kevin Duh, and Nicholas Andrews. 2025. Uncertainty distillation: Teaching language models to express semantic confidence. Preprint, arXiv:2503.14749.
  14. Kairui Hu, Penghao Wu, Fanyi Pu, Wang Xiao, Yuanhan Zhang, Xiang Yue, Bo Li, and Ziwei Liu. 2025. Video-mmmu: Evaluating knowledge acquisition from multi-discipline professional videos. Preprint, arXiv:2501.13826. — [arXiv:2501.13826](https://arxiv.org/abs/2501.13826) find
  15. Kimi Team, Angang Du, Bohong Yin, Bowei Xing, Bowen Qu, Bowen Wang, Cheng Chen, Chenlin Zhang, Chenzhuang Du, Chu Wei, Congcong Wang, Dehao Zhang, Dikang Du, Dongliang Wang, Enming Yuan, Enzhe Lu, Fang Li, Flood Sung, Guangda Wei, and 73 others. 2025. Kimi-VL technical report. Preprint, arXiv:2504.0…
  16. Lorenz Kuhn, Yarin Gal, and Sebastian Farquhar. 2023. Semantic uncertainty: Linguistic invariances for uncertainty estimation in natural language generation. Preprint, arXiv:2302.09664. find
  17. Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. Gonzalez, Hao Zhang, and Ion Stoica. 2023. Efficient memory management for large language model serving with pagedattention. In Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles.
  18. Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Peiyuan Zhang, Yanwei Li, Ziwei Liu, and 1 others. 2024a. Llava-onevision: Easy visual task transfer. arXiv preprint arXiv:2408.03326.
  19. Jiaang Li, Yova Kementchedjhieva, Constanza Fierro, and Anders Søgaard. 2024b. Do vision and language models share concepts? a vector space alignment study. Transactions of the Association for Computational Linguistics, 12:1232–1249. find
  20. Xiujun Li, Yujie Lu, Zhe Gan, Jianfeng Gao, William Yang Wang, and Yejin Choi. 2024c. Text as images: Can multimodal large language models follow printed instructions in pixels? Preprint, arXiv:2311.17647. — [arXiv:2311.17647](https://arxiv.org/abs/2311.17647) find
  21. Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. 2023. Visual instruction tuning. Advances in neural information processing systems, 36:34892– 34916.
  22. Pan Lu, Hritik Bansal, Tony Xia, Jiacheng Liu, Chunyuan Li, Hannaneh Hajishirzi, Hao Cheng, Kai-Wei Chang, Michel Galley, and Jianfeng Gao. 2024. Mathvista: Evaluating mathematical reasoning of foundation models in visual contexts. In International Conference on Learning Representations (ICLR).
  23. Marco Mistretta, Alberto Baldrati, Lorenzo Agnolucci, Marco Bertini, and Andrew D. Bagdanov. 2025. Cross the gap: Exposing the intra-modal misalignment in clip via modality inversion. Preprint, arXiv:2502.04263. — [arXiv:2502.04263](https://arxiv.org/abs/2502.04263) find
  24. Alexander Nikitin, Jannik Kossen, Yarin Gal, and Pekka Marttinen. 2024. Kernel language entropy: Fine-grained uncertainty quantification for llms from semantic similarities. Preprint, arXiv:2405.20003. — [arXiv:2405.20003](https://arxiv.org/abs/2405.20003) find
  25. OpenAI. 2025. Introducing openai o3 and o4-mini. https://openai.com/index/ introducing-o3-and-o4-mini/. ↗
  26. OpenAI, Ahmed El-Kishky, Alexander Wei, Andre Saraiva, Borys Minaiev, Daniel Selsam, David Dohan, Francis Song, Hunter Lightman, Ignasi Clavera, Jakub Pachocki, Jerry Tworek, Lorenz Kuhn, Lukasz Kaiser, Mark Chen, Max Schwarzer, Mostafa Rohaninejad, Nat McAleese, o3 contributors, and 6 others. 2025.… find
  27. OpenAI, Aaron Hurst, Adam Lerer, Adam P. Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, Aleksander M adry, Alex Baker-Whitcomb, Alex Beutel, Alex Borzunov, Alex Carney, Alex Chow, Alex Kirillov, Alex Nichol, and 400 others. 2024a. Gpt-4o sy…
  28. OpenAI, Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, Aleksander Madry, Alex Beutel, Alex Carney, Alex Iftimie, Alex Karpenko, Alex Tachard Passos, Alexander Neitz, Alexander Prokofiev, Alexander Wei, Allison Tam, Ally Bennett, and 243 others. 2024b.… — OpenAI o1 System Card
  29. Piotr Padlewski, Max Bain, Matthew Henderson, Zhongkai Zhu, Nishant Relan, Hai Pham, Donovan Ong, Kaloyan Aleksiev, Aitor Ormazabal, Samuel Phua, Ethan Yeo, Eugenie Lamprecht, Qi Liu, Yuqi Wang, Eric Chen, Deyu Fu, Lei Li, Che Zheng, Cyprien de Masson d’Autume, and 3 others. 2024. Vibe-eval: A hard…
  30. Eunkyu Park, Minyeong Kim, and Gunhee Kim. 2025. Halloc: Token-level localization of hallucinations for vision language models. Preprint, arXiv:2506.10286. — [arXiv:2506.10286](https://arxiv.org/abs/2506.10286) find
  31. Yi Peng, Chris, Xiaokun Wang, Yichen Wei, Jiangbo Pei, Weijie Qiu, Ai Jian, Yunzhuo Hao, Jiachun Pan, Tianyidan Xie, Li Ge, Rongxian Zhuang, Xuchen Song, Yang Liu, and Yahui Zhou. 2025. Skywork r1v: Pioneering multimodal reasoning with chain-of-thought. Preprint, arXiv:2504.05599.
  32. Dong Shu, Haiyan Zhao, Jingyu Hu, Weiru Liu, Ali Payani, Lu Cheng, and Mengnan Du. 2025. Large vision-language model alignment and misalignment: A survey through the lens of explainability. Preprint, arXiv:2501.01346. find
  33. Linwei Tao, Yi-Fan Yeh, Minjing Dong, Tao Huang, Philip Torr, and Chang Xu. 2025. Revisiting uncertainty estimation and calibration of large language models. arXiv preprint arXiv:2505.23854. — [arXiv:2505.23854](https://arxiv.org/abs/2505.23854) find
  34. Katherine Tian, Eric Mitchell, Allan Zhou, Archit Sharma, Rafael Rafailov, Huaxiu Yao, Chelsea Finn, and Christopher 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 Conferen… find
  35. Roman Vashurin, Ekaterina Fadeeva, Artem Vazhentsev, Lyudmila Rvanova, Daniil Vasilev, Akim Tsvigun, Sergey Petrakov, Rui Xing, Abdelrahman Sadallah, Kirill Grishchenkov, Alexander Panchenko, Timothy Baldwin, Preslav Nakov, Maxim Panov, and Artem Shelmanov. 2024. Benchmarking uncertainty quantificat…
  36. Ke Wang, Junting Pan, Weikang Shi, Zimu Lu, Houxing Ren, Aojun Zhou, Mingjie Zhan, and Hongsheng Li. 2024. Measuring multimodal mathematical reasoning with math-vision dataset. In The Thirty-eight Conference on Neural Information Processing Systems Datasets and Benchmarks Track.
  37. Peiyu Wang, Yichen Wei, Yi Peng, Xiaokun Wang, Weijie Qiu, Wei Shen, Tianyidan Xie, Jiangbo Pei, Jianhao Zhang, Yunzhuo Hao, and 1 others. 2025a. Skywork r1v2: Multimodal hybrid reinforcement learning for reasoning. arXiv preprint arXiv:2504.16656.
  38. Xiyao Wang, Zhengyuan Yang, Chao Feng, Yongyuan Liang, Yuhang Zhou, Xiaoyu Liu, Ziyi Zang, Ming Li, Chung-Ching Lin, Kevin Lin, Linjie Li, Furong Huang, and Lijuan Wang. 2025b. Vicrit: A verifiable reinforcement learning proxy task for visual perception in vlms. Preprint, arXiv:2506.10128. — [arXiv:2506.10128](https://arxiv.org/abs/2506.10128) find
  39. Yanling Wang, Yihan Zhao, Xiaodong Chen, Shasha Guo, Lixin Liu, Haoyang Li, Yong Xiao, Jing Zhang, Qi Li, and Ke Xu. 2025c. Visualsimpleqa: A benchmark for decoupled evaluation of large vision-language models in fact-seeking question answering. Preprint, arXiv:2503.06492. — [arXiv:2503.06492](https://arxiv.org/abs/2503.06492) find
  40. Miao Xiong, Zhiyuan Hu, Xinyang Lu, Yifei Li, Jie Fu, Junxian He, and Bryan Hooi. 2024. Can llms express their uncertainty? an empirical evaluation of confidence elicitation in llms. Preprint, arXiv:2306.13063. find
  41. Tianyang Xu, Shujin Wu, Shizhe Diao, Xiaoze Liu, Xingyao Wang, Yangyi Chen, and Jing Gao. 2024. find
  42. SaySelf: Teaching LLMs to express confidence with self-reflective rationales. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 5985–5998, Miami, Florida, USA. Association for Computational Linguistics. find
  43. Daniel Yang, Yao-Hung Hubert Tsai, and Makoto Yamada. 2024. On verbalized confidence scores for llms. Preprint, arXiv:2412.14737. — [arXiv:2412.14737](https://arxiv.org/abs/2412.14737) find
  44. Xiang Yue, Tianyu Zheng, Yuansheng Ni, Yubo Wang, Kai Zhang, Shengbang Tong, Yuxuan Sun, Botao Yu, Ge Zhang, Huan Sun, Yu Su, Wenhu Chen, and Graham Neubig. 2024. Mmmu-pro: A more robust multi-discipline multimodal understanding benchmark. Preprint, arXiv:2409.02813. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xuan-2025-seeing-is-believing-but-how/