Elamur Research⌕
☰
home / papers / 2025 / Benchmark Self-Evolving: A Multi-Agent Framework for Dynamic LLM Evaluation

Benchmark Self-Evolving: A Multi-Agent Framework for Dynamic LLM Evaluation

Paper · 2025 · Siyuan Wang, Zhuohan Long, Zhihao Fan et al.

Summary

Static benchmarks age quickly as LLMs improve. Their test sets may also leak into training data, which inflates scores. Rather than annotating new datasets from scratch, the authors update existing ones automatically. Their framework has three goals:

  • Scalable evaluation: new or harder questions.
  • Robust evaluation: perturbed contexts.
  • Fine-grained evaluation: questions that probe sub-abilities.

The framework uses six reframing operations. Instances are produced by a GPT-4-based multi-agent pipeline that first filters out items the backbone cannot solve and then checks each new instance twice. The authors applied it to seven general and domain-specific datasets and tested seven LLMs. Most models scored lower on the evolved instances than on the originals. Gaps between models, and between tasks for the same model, also grew wider. The authors argue this gives a more realistic view of capabilities, helps choose a model for a given application, and reduces the impact of data contamination. Because the process can be repeated, benchmarks can keep evolving alongside the models.

Key points

  • Six operations in three directions:
    • Scalable: Question Alternating and Question Complicating.
    • Robust: Context Paraphrasing, Context Noising and Polarity Reversing.
    • Fine-grained: Sub-ability Question Generation, which targets task planning, implicit knowledge identification and relevant context retrieval.
  • Four agents: the first screens instances in advance, the second generates new ones, the third verifies them, and the fourth formulates candidate options. The formulator writes a wrong answer option that the verifier must reject, which serves as a second check.
  • Data: GSM8K, CLUTRR, StrategyQA, BoolQ, MedQA, HotpotQA and SPARTQA. The authors sampled 100 instances per dataset, producing about 1,600 evolving instances. GSM8K was evolved over two iterations to show that the process can continue.
  • Models evaluated: three closed-source models (GPT-4, ChatGPT, ChatGLM) and four open-source families (LLama, Mistral, Qwen, Yi).
  • Limitations: human checking still found a small share of inaccurate generated instances. Polarity reversing on factual sets such as BoolQ can introduce counterfactual content. Code and data are available at github.com/NanshineLoong/Self-Evolving-Benchmark.

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

🎓Fudan UniversityShanghai · China167 works · 20,987 citations🏢Alibaba GroupHangzhou · China168 works · 15,386 citations🎓Tongji UniversityShanghai · China19 works · 2,162 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.Evaluation & benchmarks · 1,220 works

Cite

Cite as Siyuan Wang, Zhuohan Long, Zhihao Fan, Xuanjing Huang and Zhongyu Wei. 2025. Benchmark Self-Evolving: A Multi-Agent Framework for Dynamic LLM Evaluation. Proceedings of the 31st International Conference on Computational Linguistics.

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

@inproceedings{wang2025benchmark,
  title = {{Benchmark Self-Evolving: A Multi-Agent Framework for Dynamic LLM Evaluation}},
  author = {Siyuan Wang, Zhuohan Long, Zhihao Fan, Xuanjing Huang and Zhongyu Wei},
  booktitle = {Proceedings of the 31st International Conference on Computational Linguistics},
  year = {2025},
  url = {https://aclanthology.org/2025.coling-main.223/}
}

References

37 references, 20 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. Stella Biderman, Hailey Schoelkopf, Quentin Gregory Anthony, Herbie Bradley, Kyle O’Brien, Eric Hallahan, Mohammad Aflah Khan, Shivanshu Purohit, USVSN Sai Prashanth, Edward Raff, et al. 2023. Pythia: A suite for analyzing large language models across training and scaling. In International Conferenc… find
  3. Yupeng Chang, Xu Wang, Jindong Wang, Yuan Wu, Kaijie Zhu, Hao Chen, Linyi Yang, Xiaoyuan Yi, Cunxiang Wang, Yidong Wang, et al. 2023. A survey on evaluation of large language models. arXiv preprint arXiv:2307.03109.
  4. Wei-Lin Chiang, Zhuohan Li, Zi Lin, Ying Sheng, Zhanghao Wu, Hao Zhang, Lianmin Zheng, Siyuan Zhuang, Yonghao Zhuang, Joseph E. Gonzalez, Ion Stoica, and Eric P. Xing. 2023. Vicuna: An open-source chatbot impressing gpt-4 with 90%* chatgpt quality. find
  5. Christopher Clark, Kenton Lee, Ming-Wei Chang, Tom Kwiatkowski, Michael Collins, and Kristina Toutanova. 2019. Boolq: Exploring the surprising difficulty of natural yes/no questions. NAACL.
  6. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, et al. 2021. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168.
  7. All data utilized in our benchmark self-evolving framework are sourced from publicly available datasets. Our generated evolving instances for dynamic evaluation are also publicly released for Guanting Dong, Jinxu Zhao, Tingfeng Hui, Daichi Guo, usage and have been subjected to a thorough re-Wenlong… find
  8. Xu. 2023. Revisit input perturbation problems for llms: A unified robustness evaluation framework for noisy slot filling task. Preprint, arXiv:2310.06504. — [arXiv:2310.06504](https://arxiv.org/abs/2310.06504) find
  9. Jessica López Espejel, El Hassane Ettifouri, Mahaman Sanoussi Yahaya Alassan, El Mehdi Chouham, and Walid Dahhane. 2023. Gpt-3.5, gpt-4, or bard? evaluating llms reasoning ability in zero-shot setting and performance boosting through prompts. Natural Language Processing Journal, 5:100032. find
  10. Lizhou Fan, Wenyue Hua, Lingyao Li, Haoyang Ling, Yongfeng Zhang, and Libby Hemphill. 2023. Nphardeval: Dynamic benchmark on reasoning ability of large language models via complexity classes. arXiv preprint arXiv:2312.14890.
  11. Isabel O Gallegos, Ryan A Rossi, Joe Barrow, Md Mehrab Tanjim, Sungchul Kim, Franck Dernoncourt, Tong Yu, Ruiyi Zhang, and Nesreen K Ahmed. 2023. Bias and fairness in large language models: A survey. arXiv preprint arXiv:2309.00770.
  12. Mor Geva, Daniel Khashabi, Elad Segal, Tushar Khot, Dan Roth, and Jonathan Berant. 2021. Did aristotle use a laptop? a question answering benchmark with implicit reasoning strategies. Transactions of the Association for Computational Linguistics, 9:346– 361. find
  13. Shahriar Golchin and Mihai Surdeanu. 2023. Time travel in llms: Tracing data contamination in large language models. arXiv preprint arXiv:2308.08493.
  14. Yang Liu, Dan Iter, Yichong Xu, Shuohang Wang, Ruochen Xu, and Chenguang Zhu. 2023. Gpteval: Nlg evaluation using gpt-4 with better human alignment. arXiv preprint arXiv:2303.16634.
  15. Roshanak Mirzaee, Hossein Rajaby Faghihi, Qiang Ning, and Parisa Kordjmashidi. 2021. Spartqa:: A textual question answering benchmark for spatial reasoning. arXiv preprint arXiv:2104.05832.
  16. OpenAI. 2023. Gpt-4 technical report. arXiv:2303.08774.
  17. Preprint, Baolin Peng, Chunyuan Li, Pengcheng He, Michel Galley, and Jianfeng Gao. 2023. Instruction tuning with gpt-4. arXiv preprint arXiv:2304.03277.
  18. Pouya Pezeshkpour and Estevam Hruschka. 2023. Large language models sensitivity to the order of options in multiple-choice questions. Preprint, arXiv:2308.11483.
  19. Khaled Saab, Tao Tu, Wei-Hung Weng, Ryutaro Tanno, David Stutz, Ellery Wulczyn, Fan Zhang, Tim Strother, Chunjong Park, Elahe Vedadi, et al. 2024. Capabilities of gemini models in medicine. arXiv preprint arXiv:2404.18416. find
  20. Oscar Sainz, Jon Ander Campos, Iker García-Ferrero, Julen Etxaniz, Oier Lopez de Lacalle, and Eneko Agirre. 2023. Nlp evaluation in trouble: On the need to measure llm data contamination for each benchmark. arXiv preprint arXiv:2310.18018.
  21. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2020. Measuring massive multitask language understanding. arXiv preprint arXiv:2009.03300.
  22. Weijia Shi, Anirudh Ajith, Mengzhou Xia, Yangsibo Huang, Daogao Liu, Terra Blevins, Danqi Chen, and Luke Zettlemoyer. 2023. Detecting pretraining data from large language models. arXiv preprint arXiv:2310.16789. find
  23. Albert Q Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, et al. 2023. Mistral 7b. arXiv preprint arXiv:2310.06825. find
  24. Koustuv Sinha, Shagun Sodhani, Jin Dong, Joelle Pineau, and William L Hamilton. 2019. Clutrr: A diagnostic benchmark for inductive reasoning from text. arXiv preprint arXiv:1908.06177. find
  25. Di Jin, Eileen Pan, Nassim Oufattole, Wei-Hung Weng, Hanyi Fang, and Peter Szolovits. 2021. What disease does this patient have? a large-scale open domain question answering dataset from medical exams. Applied Sciences, 11(14):6421.
  26. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. 2023. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971.
  27. Douwe Kiela, Max Bartolo, Yixin Nie, Divyansh Kaushik, Atticus Geiger, Zhengxuan Wu, Bertie Vidgen, Grusha Prasad, Amanpreet Singh, Pratik Ringshia, et al. 2021. Dynabench: Rethinking benchmarking in nlp. arXiv preprint arXiv:2104.14337. find
  28. Siyuan Wang, Zhongyu Wei, Yejin Choi, and Xiang Ren. 2024. Can llms reason with rules? logic scaffolding for stress-testing and improving llms. arXiv preprint arXiv:2402.11442. find
  29. Dong-Ho Lee, Jay Pujara, Mohit Sewak, Ryen W White, and Sujay Kumar Jauhar. 2023. Making large language models better data creators. arXiv preprint arXiv:2310.20111. find
  30. Yizhong Wang, Yeganeh Kordi, Swaroop Mishra, Alisa Liu, Noah A Smith, Daniel Khashabi, and Hannaneh Hajishirzi. 2022. Self-instruct: Aligning language models with self-generated instructions. arXiv preprint arXiv:2212.10560. find
  31. Percy Liang, Rishi Bommasani, Tony Lee, Dimitris Tsipras, Dilara Soylu, Michihiro Yasunaga, Yian Zhang, Deepak Narayanan, Yuhuai Wu, Ananya Kumar, et al. 2022. Holistic evaluation of language models. arXiv preprint arXiv:2211.09110. find
  32. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in Neural Information Processing Systems, 35:24824–24837.
  33. Tianwen Wei, Liang Zhao, Lichang Zhang, Bo Zhu, Lijie Wang, Haihua Yang, Biye Li, Cheng Cheng, Weiwei Lü, Rui Hu, Chenxia Li, Liu Yang, Xilin Luo, Xuejie Wu, Lunan Liu, Wenjun Cheng, Peng Cheng, Jianhao Zhang, Xiaoyu Zhang, Lei Lin, Xiaokun Wang, Yutuan Ma, Chuanhai Dong, Yanqi Sun, Yifu Chen, Yongy… — Skywork: A More Open Bilingual Foundation Model
  34. Kun Zhou, Yutao Zhu, Zhipeng Chen, Wentong Chen, Wayne Xin Zhao, Xu Chen, Yankai Lin, Ji-Rong Wen, and Jiawei Han. 2023. Don’t make your llm an evaluation benchmark cheater. arXiv preprint arXiv:2311.01964. find
  35. Can Xu, Qingfeng Sun, Kai Zheng, Xiubo Geng, Pu Zhao, Jiazhan Feng, Chongyang Tao, and Daxin Jiang. 2023. Wizardlm: Empowering large language models to follow complex instructions. arXiv preprint arXiv:2304.12244.
  36. Kaijie Zhu, Jindong Wang, Qinlin Zhao, Ruochen Xu, and Xing Xie. 2024. Dyval 2: Dynamic evaluation of large language models by meta probing agents. arXiv preprint arXiv:2402.14865.
  37. An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, et al. 2024. Qwen2 technical report. arXiv preprint arXiv:2407.10671. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2025-benchmark-self-evolving-multi-agent/