---
title: "Verifying the Subjective — paper by Sha and Zhu (2026)"
description: "To extend the Reinforcement Learning with Verifiable Rewards (RLVR) idea to open-ended, subjective tasks, the authors… 58 references."
authors: ["Jiu Sha", "Mengxiao Zhu"]
year: 2026
venue: "Findings of the Association for Computational Linguistics: ACL 2026"
source: "https://aclanthology.org/2026.findings-acl.1174/"
doi: "10.18653/v1/2026.findings-acl.1174"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/sha-2026-verifying-subjective-structured-multilingual-rewards/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/sha-2026-verifying-subjective-structured-multilingual-rewards/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/47/) / Verifying the Subjective: Structured Multilingual Rewards for Low-Resource Alignment

# Verifying the Subjective: Structured Multilingual Rewards for Low-Resource Alignment

**Authors:** Jiu Sha and [Mengxiao Zhu](https://md.research.elamur.ai/free/en/authors/mengxiao-zhu/) · **Year:** 2026 · **Venue:** [Findings of the Association for Computational Linguistics: ACL 2026](https://md.research.elamur.ai/free/en/venues/findings/) ([Findings of the Association for Computational Linguistics 2026](https://md.research.elamur.ai/free/en/venues/findings/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.findings-acl.1174/) · [PDF](https://aclanthology.org/2026.findings-acl.1174.pdf) · [DOI 10.18653/v1/2026.findings-acl.1174](https://doi.org/10.18653/v1/2026.findings-acl.1174) · [OpenAlex](https://openalex.org/W7166873924)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
Aligning LLMs for low-resource languages is limited by weak reward signals. Scalar reward models transfer poorly across cultures, free-text judge critiques are noisy and hard to check, and rule-based verification only works when answers are objectively right or wrong. To extend the Reinforcement Learning with Verifiable Rewards (RLVR) idea to open-ended, subjective tasks, the authors introduce what they call a Structured Multilingual Reward Modeling Framework. It has three parts:

- **Checklist schema:** quality is split into a weighted checklist of shared reasoning criteria plus task-specific answer criteria.
- **SGCM:** a critic trained on human-written structured critiques gives each checklist item an evidence-based justification and a score between 0 and 1.
- **AMRO:** these scores are combined with language-consistency and task-recognition signals into a single reward.

The reward drives a bootstrapped GRPO pipeline with three stabilization techniques. The authors also build and release a critique dataset and benchmarks covering 50 low-resource languages and 7 subjective task categories. Fine-tuned 7B/8B models gain in reasoning, response quality and safety. RL training stays stable in cases where ordinary LLM judges cause collapse or oscillation.

## Key points
- **Base models and language check:** Llama-3.1-8B-Instruct and Qwen2.5-7B-Instruct were trained with SFT followed by RL. The language-consistency reward is a binary check that the output is in the expected language, based on Google Translate, langid and FastText. Code is at github.com/Shajiu/SGCM.
- **New benchmarks:** the main suites are Multi-50L-5C-Bench and Multi-50L-Safety-Bench. On the safety benchmark, the two SFT+RL models scored 82.31 and 91.67. The paper reports that these scores beat all open-source baselines and most API models.
- **Reward ablation (Qwen2.5-7B):**
  - SGCM+AMRO reached 47.11 on 5C-Bench and 86.99 on Safety-Bench.
  - GPT-5 (high) as judge gave 36.61 / 70.69, and Gemini-2.5-Pro gave 37.31 / 72.89; both produced unstable training.
  - A Qwen2.5-7B-Instruct judge caused collapse after 130 steps.
- **GRPO stabilization:** combining length-aware loss normalization, non-zero reward variance and SD stabilization raised scores from 39.40 / 75.84 to 47.11 / 86.99. Across five seeds, the divergence rate fell from 28% to 4%.
- **Open benchmarks (Qwen2.5-7B, base → SFT+RL):**
  - Flores-101: 69.11 → 74.28
  - Translated MMLU: 69.08 → 73.11
  - IndoMMLU: 56.42 → 61.05

*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

- [Minzu University of China](https://md.research.elamur.ai/free/en/universities/minzu-university-of-china/) — University, Beijing, China · 5 works · 2 citations

Also on the byline: [North China University of Technology](https://ror.org/01nky7652).

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

🧭 Topics (2): [🌍 Multilingual & cross-lingual NLP](https://md.research.elamur.ai/free/en/topics/multilingual-nlp/) · 828 works · [🌱 Low-resource languages](https://md.research.elamur.ai/free/en/topics/low-resource-languages/) · 267 works

## Cite

Cite as: Jiu Sha and Mengxiao Zhu. 2026. Verifying the Subjective: Structured Multilingual Rewards for Low-Resource Alignment. Findings of the Association for Computational Linguistics: ACL 2026.

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

```bibtex
@inproceedings{sha2026verifying,
  title = {{Verifying the Subjective: Structured Multilingual Rewards for Low-Resource Alignment}},
  author = {Jiu Sha and Mengxiao Zhu},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1174},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1174}
}
```

- [← previous VERICITE: Evaluating Sentence-Level Citation Faithfulness in Retrieval-Augmented Medical Question Answering](https://md.research.elamur.ai/free/en/papers/ma-2026-vericite-evaluating-sentence-level-citation/)
- [next → VerilogLAVD: LLM-Aided Pattern Generation for Verilog CWE Detection](https://md.research.elamur.ai/free/en/papers/long-2026-veriloglavd-llm-aided-pattern-generation/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/47/)

## References

58 references, 28 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, Harkirat Behl, Sébastien Bubeck, Ronen Eldan, Suriya Gunasekar, Michael Harrison, Russell J Hewett, Mojan Javaheripi, Piero Kauffmann, and 1 others. 2024. [Phi-4 technical report](https://md.research.elamur.ai/free/en/papers/abdin-2024-phi-4-technical-report/). arXiv preprint arXiv:2412.08905.
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](https://md.research.elamur.ai/free/en/papers/openai-2023-gpt-4-technical-report/). arXiv preprint arXiv:2303.08774.
3. Arthur Mensch Chris Bamford Devendra Singh Chaplot Diego de las Casas Florian Bressand Gianna Lengyel Guillaume Lample Lucile Saulnier Lélio Renard Lavaud Marie-Anne Lachaux Pierre Stock Teven Le Scao Thibaut Lavril Thomas Wang Timothée Lacroix William El Sayed Albert Q. Jiang, Alexandre Sablayrolle… [find](https://scholar.google.com/scholar?q=Arthur+Mensch+Chris+Bamford+Devendra+Singh+Chaplot+Diego+de+las+Casas+Florian+Bressand+Gianna+Lengyel+Guillaume+Lample+Lucile+Saulnier+L%C3%A9lio+Renard+Lavaud+Marie%2DAnne+Lachaux+Pierre+Stock+Teven+Le+Scao)
4. Anthropic. 2025. Introducing claude sonnet 4.5. Journal of Clinical Oncology, 33(27):2025–10–21. [find](https://scholar.google.com/scholar?q=Anthropic%2E+2025%2E+Introducing+claude+sonnet+4%2E5%2E+Journal+of+Clinical+Oncology%2C+33%2827%29%3A2025%E2%80%9310%E2%80%9321%2E)
5. Benjamin Samuel Bloom. 2010. A taxonomy for learning, teaching, and assessing: A revision of Bloom’s taxonomy of educational objectives. Longman. [find](https://scholar.google.com/scholar?q=Benjamin+Samuel+Bloom%2E+2010%2E+A+taxonomy+for+learning%2C+teaching%2C+and+assessing%3A+A+revision+of+Bloom%E2%80%99s+taxonomy+of+educational+objectives%2E+Longman%2E)
6. Zhihong Chen, Shuo Yan, Juhao Liang, Feng Jiang, Xiangbo Wu, Fei Yu, Guiming Hardy Chen, Junying Chen, Hongbo Zhang, Li Jianquan, Wan Xiang, and Benyou Wang. 2023. MultilingualSIFT: Multilingual Supervised Instruction Fine-tuning. [find](https://scholar.google.com/scholar?q=Zhihong+Chen%2C+Shuo+Yan%2C+Juhao+Liang%2C+Feng+Jiang%2C+Xiangbo+Wu%2C+Fei+Yu%2C+Guiming+Hardy+Chen%2C+Junying+Chen%2C+Hongbo+Zhang%2C+Li+Jianquan%2C+Wan+Xiang%2C+and+Benyou+Wang%2E+2023%2E+MultilingualSIFT%3A+Multilingual+Super)
7. Tianzhe Chu, Yuexiang Zhai, Jihan Yang, Shengbang Tong, Saining Xie, Dale Schuurmans, Quoc V Le, Sergey Levine, and Yi Ma. 2025a. [Sft memorizes, rl generalizes: A comparative study of foundation model post-training](https://md.research.elamur.ai/free/en/papers/chu-2025-sft-memorizes-rl-generalizes-comparative/). arXiv preprint arXiv:2501.17161.
8. Xiangxiang Chu, Hailang Huang, Xiao Zhang, Fei Wei, and Yong Wang. 2025b. Gpg: A simple and strong reinforcement learning baseline for model reasoning. arXiv preprint arXiv:2504.02546. — [arXiv:2504.02546](https://arxiv.org/abs/2504.02546) [find](https://scholar.google.com/scholar?q=Xiangxiang+Chu%2C+Hailang+Huang%2C+Xiao+Zhang%2C+Fei+Wei%2C+and+Yong+Wang%2E+2025b%2E+Gpg%3A+A+simple+and+strong+reinforcement+learning+baseline+for+model+reasoning%2E+arXiv+preprint+arXiv%3A2504%2E02546%2E+%E2%80%94+%5BarXiv%3A2504%2E0)
9. 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](https://md.research.elamur.ai/free/en/papers/clark-2018-think-have-solved-question-answering/). ArXiv, abs/1803.05457.
10. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, and 1 others. 2025. [Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities](https://md.research.elamur.ai/free/en/papers/comanici-2025-gemini-2-5-pushing-frontier/). arX…
11. DeepSeek-AI. 2024. Deepseek-v3 technical report. [find](https://scholar.google.com/scholar?q=DeepSeek%2DAI%2E+2024%2E+Deepseek%2Dv3+technical+report%2E)
12. Michael Desmond, Zahra Ashktorab, Werner Geyer, Elizabeth M. Daly, Martín Santillán Cooper, Qian Pan, Rahul Nair, Nico Wagner, and Tejaswini Pedapati. 2025. Evalassist: Llm-as-a-judge simplified. In AAAI-25, Sponsored by the Association for the Advancement of Artificial Intelligence, February 25 -Ma… [find](https://scholar.google.com/scholar?q=Michael+Desmond%2C+Zahra+Ashktorab%2C+Werner+Geyer%2C+Elizabeth+M%2E+Daly%2C+Mart%C3%ADn+Santill%C3%A1n+Cooper%2C+Qian+Pan%2C+Rahul+Nair%2C+Nico+Wagner%2C+and+Tejaswini+Pedapati%2E+2025%2E+Evalassist%3A+Llm%2Das%2Da%2Djudge+simplified%2E+In+A)
13. 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](https://md.research.elamur.ai/free/en/papers/grattafiori-2024-llama-3-herd-models/). arXiv e-prints, pages arXiv–2407.
14. Olga Majewska Qianchu Liu Ivan Vuli’c Edoardo M. Ponti, Goran Glava s and Anna Korhonen. 2020. [XCOPA: A multilingual dataset for causal commonsense reasoning](https://md.research.elamur.ai/free/en/papers/ponti-2020-xcopa-multilingual-dataset-causal-commonsense/). arXiv preprint.
15. Aryo Pradipta Gema, Joshua Ong Jun Leang, Giwon Hong, Alessio Devoto, Alberto Carlo Maria Mancino, Rohit Saxena, Xuanli He, Yu Zhao, Xiaotang Du, Mohammad Reza Ghasemi Madani, Claire Barale, Robert McHardy, Joshua Harris, Jean Kaddour, Emile van Krieken, and Pasquale Minervini. 2024. Are we done wit… [find](https://scholar.google.com/scholar?q=Aryo+Pradipta+Gema%2C+Joshua+Ong+Jun+Leang%2C+Giwon+Hong%2C+Alessio+Devoto%2C+Alberto+Carlo+Maria+Mancino%2C+Rohit+Saxena%2C+Xuanli+He%2C+Yu+Zhao%2C+Xiaotang+Du%2C+Mohammad+Reza+Ghasemi+Madani%2C+Claire+Barale%2C+Robert+Mc)
16. Team GLM, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, Diego Rojas, Guanyu Feng, Hanlin Zhao, Hanyu Lai, Hao Yu, Hongning Wang, Jiadai Sun, Jiajie Zhang, Jiale Cheng, Jiayi Gui, Jie Tang, Jing Zhang, Juanzi Li, and 37 others. 2024. [Chatglm: A family of large language models from glm-130b t…](https://md.research.elamur.ai/free/en/papers/glm-2024-chatglm-family-large-language-models/)
17. Naman Goyal, Cynthia Gao, Vishrav Chaudhary, Peng-Jen Chen, Guillaume Wenzek, Da Ju, Sanjana Krishnan, Marc’Aurelio Ranzato, Francisco Guzmán, and Angela Fan. 2021. The flores-101 evaluation benchmark for low-resource and multilingual machine translation. [find](https://scholar.google.com/scholar?q=Naman+Goyal%2C+Cynthia+Gao%2C+Vishrav+Chaudhary%2C+Peng%2DJen+Chen%2C+Guillaume+Wenzek%2C+Da+Ju%2C+Sanjana+Krishnan%2C+Marc%E2%80%99Aurelio+Ranzato%2C+Francisco+Guzm%C3%A1n%2C+and+Angela+Fan%2E+2021%2E+The+flores%2D101+evaluation+benchmark)
18. 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](https://md.research.elamur.ai/free/en/papers/deepseek-ai-2025-deepseek-r1-incentivizing-reasoning-capability/). arXiv preprint arXiv:2501.12948.
19. Dan Hendrycks, Collin Burns, Steven Basart, Andrew Critch, Jerry Li, Dawn Song, and Jacob Steinhardt. 2021. [Aligning ai with shared human values](https://md.research.elamur.ai/free/en/papers/hendrycks-2020-aligning-ai-shared-human-values/). Proceedings of the International Conference on Learning Representations (ICLR).
20. Xia Hou, Qifeng Li, Jian Yang, Tongliang Li, Linzheng Chai, Xianjie Wu, Hangyuan Ji, Zhoujun Li, Jixuan Nie, Jingbo Dun, and 1 others. 2024. Raw text is all you need: Knowledge-intensive multi-turn instruction tuning for large language model. arXiv preprint arXiv:2407.03040. — [arXiv:2407.03040](https://arxiv.org/abs/2407.03040) [find](https://scholar.google.com/scholar?q=Xia+Hou%2C+Qifeng+Li%2C+Jian+Yang%2C+Tongliang+Li%2C+Linzheng+Chai%2C+Xianjie+Wu%2C+Hangyuan+Ji%2C+Zhoujun+Li%2C+Jixuan+Nie%2C+Jingbo+Dun%2C+and+1+others%2E+2024%2E+Raw+text+is+all+you+need%3A+Knowledge%2Dintensive+multi%2Dturn+in)
21. Jaedong Hwang, Kumar Tanmay, Seok-Jin Lee, Ayush Agrawal, Hamid Palangi, Kumar Ayush, Ila Fiete, and Paul Pu Liang. 2025. Learn globally, speak locally: Bridging the gaps in multilingual reasoning. arXiv preprint arXiv:2507.05418. — [arXiv:2507.05418](https://arxiv.org/abs/2507.05418) [find](https://scholar.google.com/scholar?q=Jaedong+Hwang%2C+Kumar+Tanmay%2C+Seok%2DJin+Lee%2C+Ayush+Agrawal%2C+Hamid+Palangi%2C+Kumar+Ayush%2C+Ila+Fiete%2C+and+Paul+Pu+Liang%2E+2025%2E+Learn+globally%2C+speak+locally%3A+Bridging+the+gaps+in+multilingual+reasoning%2E+ar)
22. Armand Joulin, Edouard Grave, Piotr Bojanowski, Matthijs Douze, Hérve Jégou, and Tomas Mikolov. 2016. [Fasttext](https://md.research.elamur.ai/free/en/papers/joulin-2016-fasttext-zip-compressing-text-classification/). zip: Compressing text classification models. arXiv preprint arXiv:1612.03651.
23. Fajri Koto, Nurul Aisyah, Haonan Li, and Timothy Baldwin. 2023. Large language models only pass primary school exams in Indonesia: A comprehensive test on IndoMMLU. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP), Singapore. Association for Computati… [find](https://scholar.google.com/scholar?q=Fajri+Koto%2C+Nurul+Aisyah%2C+Haonan+Li%2C+and+Timothy+Baldwin%2E+2023%2E+Large+language+models+only+pass+primary+school+exams+in+Indonesia%3A+A+comprehensive+test+on+IndoMMLU%2E+In+Proceedings+of+the+2023+Conferen)
24. Nathan Lambert, Jacob Morrison, Valentina Pyatkin, Shengyi Huang, Hamish Ivison, Faeze Brahman, Lester James V Miranda, Alisa Liu, Nouha Dziri, Shane Lyu, and 1 others. 2024. [T\\" ulu 3: Pushing frontiers in open language model post-training](https://md.research.elamur.ai/free/en/papers/lambert-2024-tulu-3-pushing-frontiers-open/). arXiv preprint arXiv:2411.15124.
25. Stephanie Lin, Jacob Hilton, and Owain Evans. 2022. [Truthfulqa: Measuring how models mimic human falsehoods](https://md.research.elamur.ai/free/en/papers/lin-2022-truthfulqa-measuring-how-models-mimic/). Preprint, arXiv:2109.07958.
26. Chris Yuhao Liu, Liang Zeng, Jiacai Liu, Rui Yan, Jujie He, Chaojie Wang, Shuicheng Yan, Yang Liu, and Yahui Zhou. 2024. [Skywork-reward: Bag of tricks for reward modeling in llms](https://md.research.elamur.ai/free/en/papers/liu-2024-skywork-reward-bag-tricks/). arXiv preprint arXiv:2410.18451.
27. Yang Liu, Dan Iter, Yichong Xu, Shuohang Wang, Ruochen Xu, and Chenguang Zhu. 2023. [G-eval: Nlg evaluation using gpt-4 with better human alignment](https://md.research.elamur.ai/free/en/papers/liu-2023-g-eval-nlg-evaluation-gpt/) (2023). arXiv preprint arXiv:2303.16634, 12.
28. Zichen Liu, Changyu Chen, Wenjun Li, Penghui Qi, Tianyu Pang, Chao Du, Wee Sun Lee, and Min Lin. 2025. [Understanding r1-zero-like training: A critical perspective](https://md.research.elamur.ai/free/en/papers/liu-2025-understanding-r1-zero-like-training/). arXiv preprint arXiv:2503.20783.
29. Marco Lui and Timothy Baldwin. 2012. langid. py: An off-the-shelf language identification tool. In Proceedings of the ACL 2012 system demonstrations, pages 25–30. — [langid.py: An Off-the-shelf Language Identification Tool](https://md.research.elamur.ai/free/en/papers/lui-2012-langid-py-off-shelf-language/)
30. TQ Luong, X Zhang, Z Jie, P Sun, X Jin, and H Reft Li. 2024. [Reasoning with reinforced fine-tuning](https://md.research.elamur.ai/free/en/papers/trung-2024-reft-reasoning-reinforced-fine-tuning/). arXiv preprint arXiv:2401.08967.
31. Yiran Ma, Zui Chen, Tianqiao Liu, Mi Tian, Zhuo Liu, Zitao Liu, and Weiqi Luo. 2025. What are step-level reward models rewarding? counterintuitive findings from mcts-boosted mathematical reasoning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 24812–24820. [find](https://scholar.google.com/scholar?q=Yiran+Ma%2C+Zui+Chen%2C+Tianqiao+Liu%2C+Mi+Tian%2C+Zhuo+Liu%2C+Zitao+Liu%2C+and+Weiqi+Luo%2E+2025%2E+What+are+step%2Dlevel+reward+models+rewarding%3F+counterintuitive+findings+from+mcts%2Dboosted+mathematical+reasoning%2E+In)
32. Niklas Muennighoff, Thomas Wang, Lintang Sutawika, Adam Roberts, Stella Biderman, Teven Le Scao, M Saiful Bari, Sheng Shen, Zheng-Xin Yong, Hailey Schoelkopf, Xiangru Tang, Dragomir Radev, Alham Fikri Aji, Khalid Almubarak, Samuel Albanie, Zaid Alyafeai, Albert Webson, Edward Raff, and Colin Raffel.… — [Crosslingual Generalization through Multitask Finetuning](https://md.research.elamur.ai/free/en/papers/muennighoff-2022-crosslingual-generalization-through-multitask-finetuning/)
33. Mihir Prabhudesai, Lili Chen, Alex Ippoliti, Katerina Fragkiadaki, Hao Liu, and Deepak Pathak. 2025. Maximizing confidence alone improves reasoning. CoRR, abs/2505.22660. — [arXiv:2505.22660](https://arxiv.org/abs/2505.22660) [find](https://scholar.google.com/scholar?q=Mihir+Prabhudesai%2C+Lili+Chen%2C+Alex+Ippoliti%2C+Katerina+Fragkiadaki%2C+Hao+Liu%2C+and+Deepak+Pathak%2E+2025%2E+Maximizing+confidence+alone+improves+reasoning%2E+CoRR%2C+abs%2F2505%2E22660%2E+%E2%80%94+%5BarXiv%3A2505%2E22660%5D%28https%3A%2F%2F)
34. Valentina Pyatkin, Saumya Malik, Victoria Graf, Hamish Ivison, Shengyi Huang, Pradeep Dasigi, Nathan Lambert, and Hannaneh Hajishirzi. 2025. [Generalizing verifiable instruction following](https://md.research.elamur.ai/free/en/papers/pyatkin-2025-generalizing-verifiable-instruction-following/). arXiv preprint arXiv:2507.02833.
35. Haoran Que, Feiyu Duan, Liqun He, Yutao Mou, Wangchunshu Zhou, Jiaheng Liu, Wenge Rong, Zekun Moore Wang, Jian Yang, Ge Zhang, and 1 others. 2024. Hellobench: Evaluating long text generation capabilities of large language models. arXiv preprint arXiv:2409.16191. — [arXiv:2409.16191](https://arxiv.org/abs/2409.16191) [find](https://scholar.google.com/scholar?q=Haoran+Que%2C+Feiyu+Duan%2C+Liqun+He%2C+Yutao+Mou%2C+Wangchunshu+Zhou%2C+Jiaheng+Liu%2C+Wenge+Rong%2C+Zekun+Moore+Wang%2C+Jian+Yang%2C+Ge+Zhang%2C+and+1+others%2E+2024%2E+Hellobench%3A+Evaluating+long+text+generation+capabilit)
36. Abhinav Rastogi, Albert Q Jiang, Andy Lo, Gabrielle Berrada, Guillaume Lample, Jason Rute, Joep Barmentlo, Karmesh Yadav, Kartik Khandelwal, Khyathi Raghavi Chandu, and 1 others. 2025. Magistral. arXiv preprint arXiv:2506.10910. [find](https://scholar.google.com/scholar?q=Abhinav+Rastogi%2C+Albert+Q+Jiang%2C+Andy+Lo%2C+Gabrielle+Berrada%2C+Guillaume+Lample%2C+Jason+Rute%2C+Joep+Barmentlo%2C+Karmesh+Yadav%2C+Kartik+Khandelwal%2C+Khyathi+Raghavi+Chandu%2C+and+1+others%2E+2025%2E+Magistral%2E+arXi)
37. Jie Ruan, Wenqing Wang, and Xiaojun Wan. 2024. Defining and detecting vulnerability in human evaluation guidelines: A preliminary study towards reliable nlg evaluation. arXiv preprint arXiv:2406.07935. [find](https://scholar.google.com/scholar?q=Jie+Ruan%2C+Wenqing+Wang%2C+and+Xiaojun+Wan%2E+2024%2E+Defining+and+detecting+vulnerability+in+human+evaluation+guidelines%3A+A+preliminary+study+towards+reliable+nlg+evaluation%2E+arXiv+preprint+arXiv%3A2406%2E07935)
38. Jiu Sha, Mengxiao Zhu, Chong Feng, and Yuming Shang. 2025. Veef-multi-llm: Effective vocabulary expansion and parameter efficient finetuning towards multilingual large language models. In Proceedings of the 31st International Conference on Computational Linguistics, pages 7963–7981. [find](https://scholar.google.com/scholar?q=Jiu+Sha%2C+Mengxiao+Zhu%2C+Chong+Feng%2C+and+Yuming+Shang%2E+2025%2E+Veef%2Dmulti%2Dllm%3A+Effective+vocabulary+expansion+and+parameter+efficient+finetuning+towards+multilingual+large+language+models%2E+In+Proceedings+)
39. Shreya Shankar, JD Zamfirescu-Pereira, Björn Hartmann, Aditya Parameswaran, and Ian Arawjo. 2024. Who validates the validators? aligning llm-assisted evaluation of llm outputs with human preferences. In Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology, pages 1–1… [find](https://scholar.google.com/scholar?q=Shreya+Shankar%2C+JD+Zamfirescu%2DPereira%2C+Bj%C3%B6rn+Hartmann%2C+Aditya+Parameswaran%2C+and+Ian+Arawjo%2E+2024%2E+Who+validates+the+validators%3F+aligning+llm%2Dassisted+evaluation+of+llm+outputs+with+human+preferences%2E+)
40. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, and 1 others. 2024. [Deepseekmath: Pushing the limits of mathematical reasoning in open language models](https://md.research.elamur.ai/free/en/papers/shao-2024-deepseekmath-pushing-limits-mathematical-reasoning/). arXiv preprint arXiv:2402.03300.
41. Freda Shi, Mirac Suzgun, Markus Freitag, Xuezhi Wang, Suraj Srivats, Soroush Vosoughi, Hyung Won Chung, Yi Tay, Sebastian Ruder, Denny Zhou, Dipanjan Das, and Jason Wei. 2022. [Language models are multilingual chain-of-thought reasoners](https://md.research.elamur.ai/free/en/papers/shi-2022-language-models-multilingual-chain-thought/). Preprint, arXiv:2210.03057.
42. Yi Su, Dian Yu, Linfeng Song, Juntao Li, Haitao Mi, Zhaopeng Tu, Min Zhang, and Dong Yu. 2025. Crossing the reward bridge: Expanding rl with verifiable rewards across diverse domains. arXiv e-prints, pages arXiv–2503. [find](https://scholar.google.com/scholar?q=Yi+Su%2C+Dian+Yu%2C+Linfeng+Song%2C+Juntao+Li%2C+Haitao+Mi%2C+Zhaopeng+Tu%2C+Min+Zhang%2C+and+Dong+Yu%2E+2025%2E+Crossing+the+reward+bridge%3A+Expanding+rl+with+verifiable+rewards+across+diverse+domains%2E+arXiv+e%2Dprints%2C+)
43. Mirac Suzgun, Nathan Scales, Nathanael Schärli, Sebastian Gehrmann, Yi Tay, Hyung Won Chung, Aakanksha Chowdhery, Quoc V Le, Ed H Chi, Denny Zhou, , and Jason Wei. 2022. [Challenging big-bench tasks and whether chain-of-thought can solve them](https://md.research.elamur.ai/free/en/papers/suzgun-2023-challenging-big-bench-tasks-whether/). arXiv preprint arXiv:2210.09261.
44. Yongding Tao, Tian Wang, Yihong Dong, Huanyu Liu, Kechi Zhang, Xiaolong Hu, and Ge Li. 2025. Detecting data contamination from reinforcement learning post-training for large language models. arXiv preprint arXiv:2510.09259. — [arXiv:2510.09259](https://arxiv.org/abs/2510.09259) [find](https://scholar.google.com/scholar?q=Yongding+Tao%2C+Tian+Wang%2C+Yihong+Dong%2C+Huanyu+Liu%2C+Kechi+Zhang%2C+Xiaolong+Hu%2C+and+Ge+Li%2E+2025%2E+Detecting+data+contamination+from+reinforcement+learning+post%2Dtraining+for+large+language+models%2E+arXiv+pre)
45. ByteDance Seed Team. 2025. [Seed1.5-vl technical report](https://md.research.elamur.ai/free/en/papers/guo-2025-seed1-5-vl-technical/). arXiv preprint arXiv:2505.07062.
46. Qwen Team and 1 others. 2024. Qwen2 technical report. arXiv preprint arXiv:2407.10671, 2(8). [find](https://scholar.google.com/scholar?q=Qwen+Team+and+1+others%2E+2024%2E+Qwen2+technical+report%2E+arXiv+preprint+arXiv%3A2407%2E10671%2C+2%288%29%2E)
47. TIGER-Lab. 2024. Mmlu-pro (revision efb53c9). [find](https://scholar.google.com/scholar?q=TIGER%2DLab%2E+2024%2E+Mmlu%2Dpro+%28revision+efb53c9%29%2E)
48. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. [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.
49. Wei Xiong, Chenlu Ye, Baohao Liao, Hanze Dong, Xinxing Xu, Christof Monz, Jiang Bian, Nan Jiang, and Tong Zhang. 2025. Reinforce-ada: An adaptive sampling framework for reinforce-style llm training. arXiv preprint arXiv:2510.04996. — [arXiv:2510.04996](https://arxiv.org/abs/2510.04996) [find](https://scholar.google.com/scholar?q=Wei+Xiong%2C+Chenlu+Ye%2C+Baohao+Liao%2C+Hanze+Dong%2C+Xinxing+Xu%2C+Christof+Monz%2C+Jiang+Bian%2C+Nan+Jiang%2C+and+Tong+Zhang%2E+2025%2E+Reinforce%2Dada%3A+An+adaptive+sampling+framework+for+reinforce%2Dstyle+llm+training%2E+a)
50. Yixuan Even Xu, Yash Savani, Fei Fang, and J Zico Kolter. 2025. Not all rollouts are useful: Downsampling rollouts in llm reinforcement learning. arXiv preprint arXiv:2504.13818. — [arXiv:2504.13818](https://arxiv.org/abs/2504.13818) [find](https://scholar.google.com/scholar?q=Yixuan+Even+Xu%2C+Yash+Savani%2C+Fei+Fang%2C+and+J+Zico+Kolter%2E+2025%2E+Not+all+rollouts+are+useful%3A+Downsampling+rollouts+in+llm+reinforcement+learning%2E+arXiv+preprint+arXiv%3A2504%2E13818%2E+%E2%80%94+%5BarXiv%3A2504%2E13818%5D%28)
51. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 2025. [Qwen3 technical report](https://md.research.elamur.ai/free/en/papers/yang-2025-qwen3-technical-report/). arXiv preprint arXiv:2505.09388.
52. Chenlu Ye, Zhou Yu, Ziji Zhang, Hao Chen, Narayanan Sadagopan, Jing Huang, Tong Zhang, and Anurag Beniwal. 2025. Beyond correctness: Harmonizing process and outcome rewards through rl training. arXiv preprint arXiv:2509.03403. — [arXiv:2509.03403](https://arxiv.org/abs/2509.03403) [find](https://scholar.google.com/scholar?q=Chenlu+Ye%2C+Zhou+Yu%2C+Ziji+Zhang%2C+Hao+Chen%2C+Narayanan+Sadagopan%2C+Jing+Huang%2C+Tong+Zhang%2C+and+Anurag+Beniwal%2E+2025%2E+Beyond+correctness%3A+Harmonizing+process+and+outcome+rewards+through+rl+training%2E+arXiv+)
53. Rowan Zellers, Ari Holtzman, Yonatan Bisk, Ali Farhadi, and Yejin Choi. 2019. Hellaswag: Can a machine really finish your sentence? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Rowan+Zellers%2C+Ari+Holtzman%2C+Yonatan+Bisk%2C+Ali+Farhadi%2C+and+Yejin+Choi%2E+2019%2E+Hellaswag%3A+Can+a+machine+really+finish+your+sentence%3F+In+Proceedings+of+the+57th+Annual+Meeting+of+the+Association+for+Com)
54. Lunjun Zhang, Arian Hosseini, Hritik Bansal, Mehran Kazemi, Aviral Kumar, and Rishabh Agarwal. 2024. [Generative verifiers: Reward modeling as next-token prediction](https://md.research.elamur.ai/free/en/papers/zhang-2024-generative-verifiers-reward-modeling-as/). arXiv preprint arXiv:2408.15240.
55. Xiaoying Zhang, Hao Sun, Yipeng Zhang, Kaituo Feng, Chaochao Lu, Chao Yang, and Helen Meng. 2025a. Critique-grpo: Advancing LLM reasoning with natural language and numerical feedback. CoRR, abs/2506.03106. — [arXiv:2506.03106](https://arxiv.org/abs/2506.03106) [find](https://scholar.google.com/scholar?q=Xiaoying+Zhang%2C+Hao+Sun%2C+Yipeng+Zhang%2C+Kaituo+Feng%2C+Chaochao+Lu%2C+Chao+Yang%2C+and+Helen+Meng%2E+2025a%2E+Critique%2Dgrpo%3A+Advancing+LLM+reasoning+with+natural+language+and+numerical+feedback%2E+CoRR%2C+abs%2F2506%2E0)
56. Xiaoying Zhang, Hao Sun, Yipeng Zhang, Kaituo Feng, Chaochao Lu, Chao Yang, and Helen Meng. 2025b. Critique-grpo: Advancing llm reasoning with natural language and numerical feedback. arXiv preprint arXiv:2506.03106. — [arXiv:2506.03106](https://arxiv.org/abs/2506.03106) [find](https://scholar.google.com/scholar?q=Xiaoying+Zhang%2C+Hao+Sun%2C+Yipeng+Zhang%2C+Kaituo+Feng%2C+Chaochao+Lu%2C+Chao+Yang%2C+and+Helen+Meng%2E+2025b%2E+Critique%2Dgrpo%3A+Advancing+llm+reasoning+with+natural+language+and+numerical+feedback%2E+arXiv+preprint+a)
57. Yi-Fan Zhang, Tao Yu, Haochen Tian, Chaoyou Fu, Peiyan Li, Jianshu Zeng, Wulin Xie, Yang Shi, Huanyu Zhang, Junkang Wu, and 1 others. 2025c. Mm-rlhf: The next step forward in multimodal llm alignment. arXiv preprint arXiv:2502.10391. — [arXiv:2502.10391](https://arxiv.org/abs/2502.10391) [find](https://scholar.google.com/scholar?q=Yi%2DFan+Zhang%2C+Tao+Yu%2C+Haochen+Tian%2C+Chaoyou+Fu%2C+Peiyan+Li%2C+Jianshu+Zeng%2C+Wulin+Xie%2C+Yang+Shi%2C+Huanyu+Zhang%2C+Junkang+Wu%2C+and+1+others%2E+2025c%2E+Mm%2Drlhf%3A+The+next+step+forward+in+multimodal+llm+alignment%2E)
58. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, and 1 others. 2023. [Judging llm-as-a-judge with mt-bench and chatbot arena](https://md.research.elamur.ai/free/en/papers/zheng-2023-judging-llm-judge-mt-bench/). Advances in neural information processing systems, 36:46595–46623.

*Updated 1 November 2026*