Elamur Research⌕
☰
home / papers / 2025 / Skywork-R1V3 Technical Report

Skywork-R1V3 Technical Report

Paper · 2025 · Wei Shen, Jiangbo Pei, Yi Peng et al.

Summary

Open-source vision-language models still trail proprietary systems on complex visual reasoning, partly because high-quality multimodal annotations are scarce. Skywork-R1V3 tackles this by porting reasoning behaviour learned by text-only LLMs into a VLM purely through post-training, with no extra continued pre-training. Starting from InternVL-38B, the authors first apply a cold-start SFT on long chain-of-thought traces produced by their earlier Skywork-R1V2. They then run GRPO-based reinforcement learning on K12 multimodal math problems, using a combined format and accuracy reward. A final stage tunes only the connector module on multi-domain data. Two analytical findings stand out. First, the connector proves central to keeping vision and text aligned during RL. Second, the entropy of critical reasoning tokens, such as "Wait", serves as a practical signal for choosing good RL checkpoints. The resulting 38B model reaches 76.0% on MMMU, comparable to entry-level human experts and competitive with leading closed VLMs. The work matters because it shows RL can lift open VLMs substantially, and that math-focused training generalizes to other subjects.

Key points

  • Cold-start data: Skywork-R1V2 generated reasoning for about 20K Chinese high-school science questions in physics, chemistry, biology and math. Rejection sampling reduced these to roughly 12k verified examples. R1V2 itself joins InternViT-6B-448px-V2.5 with QwQ-32B.
  • RL data: 15K multimodal math items at K12 level, in multiple-choice and fill-in-the-blank formats, with answers but no reasoning steps.
  • Connector-only tuning: 10K examples spanning 20 domains, from the natural sciences through law and the arts. This stage rebalances knowledge without hurting reasoning.
  • Reward: the final reward weights accuracy at 0.8 and format at 0.2. Accuracy is checked either by an LLM verifier (e.g., Qwen2.5-Instruct 32B) or by rules.
  • RL setup: training uses GRPO in the VERL framework, with context length growing from 4,096 to 8,192. Each episode has 1,024 rollouts, 32 per prompt, with a clip range of 0.2 and no KL penalty.

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

 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 2Research subjects and languages the work is about.Reasoning · 1,426 worksVision & language · 657 works

Cite

Cite as Wei Shen, Jiangbo Pei, Yi Peng, Xuchen Song, Yang Liu, Jian Peng, Haofeng Sun, Yunzhuo Hao, Peiyu Wang, Jianhao Zhang and Yahui Zhou. 2025. Skywork-R1V3 Technical Report. arXiv:2507.06167.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2507.06167 · arXiv 2507.06167

@misc{shen2025skywork,
  title = {{Skywork-R1V3 Technical Report}},
  author = {Wei Shen, Jiangbo Pei, Yi Peng, Xuchen Song, Yang Liu, Jian Peng, Haofeng Sun, Yunzhuo Hao, Peiyu Wang, Jianhao Zhang and Yahui Zhou},
  year = {2025},
  doi = {10.48550/arxiv.2507.06167},
  eprint = {2507.06167},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2507.06167}
}

References

50 references, 23 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Anthropic. Claude-3.7, 2024. URL https://www.anthropic.com/news/claude-3-7-sonnet. ↗
  2. Lin Chen, Jinsong Li, Xiaoyi Dong, Pan Zhang, Yuhang Zang, Zehui Chen, Haodong Duan, Jiaqi Wang, Yu Qiao, Dahua Lin, et al. Are we on the right way for evaluating large vision-language models? arXiv preprint arXiv:2403.20330, 2024.
  3. Ganqu Cui, Yuchen Zhang, Jiacheng Chen, Lifan Yuan, Zhi Wang, Yuxin Zuo, Haozhan Li, Yuchen Fan, Huayu Chen, Weize Chen, Zhiyuan Liu, Hao Peng, Lei Bai, Wanli Ouyang, Yu Cheng, Bowen Zhou, and Ning Ding. The entropy mechanism of reinforcement learning for reasoning language models, 2025. URL https:/… find
  4. GOOGLE DEEPMIND. Gemini 2.5: Our most intelligent ai modeld. 2025. URL https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/\#gemini-2-5-thi nking. ↗
  5. DeepSeek-AI. Deepseek-v3 technical report, 2024. URL https://arxiv.org/abs/2412.19437. ↗
  6. DeepSeek-AI. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning, 2025. URL https://arxiv.org/abs/2501.12948. ↗
  7. Haodong Duan, Junming Yang, Yuxuan Qiao, Xinyu Fang, Lin Chen, Yuan Liu, Xiaoyi Dong, Yuhang Zang, Pan Zhang, Jiaqi Wang, et al. Vlmevalkit: An open-source toolkit for evaluating large multimodality models. In Proceedings of the 32nd ACM International Conference on Multimedia, pp. 11198–11201, 2024. find
  8. Tianrui Guan, Fuxiao Liu, Xiyang Wu, Ruiqi Xian, Zongxia Li, Xiaoyu Liu, Xijun Wang, Lichang Chen, Furong Huang, Yaser Yacoob, Dinesh Manocha, and Tianyi Zhou. Hallusionbench: An advanced diagnostic suite for entangled language hallucination and visual illusion in large vision-language models, 2024.…
  9. Dong Guo, Faming Wu, Feida Zhu, Fuxing Leng, Guang Shi, Haobin Chen, Haoqi Fan, et al. Seed1.5-vl technical report, 2025. URL https://arxiv.org/abs/2505.07062. ↗
  10. Yunzhuo Hao, Jiawei Gu, Huichen Will Wang, Linjie Li, Zhengyuan Yang, Lijuan Wang, and Yu Cheng. Can mllms reason in multimodality? emma: An enhanced multimodal reasoning benchmark. arXiv preprint arXiv:2501.05444, 2025. — [arXiv:2501.05444](https://arxiv.org/abs/2501.05444) find
  11. Jingcheng Hu, Yinmin Zhang, Qi Han, Daxin Jiang, Xiangyu Zhang, and Heung-Yeung Shum. Open-reasoner-zero: An open source approach to scaling up reinforcement learning on the base model, 2025. URL https://arxiv.org/abs/2503.24290. ↗
  12. Minyoung Huh, Brian Cheung, Tongzhou Wang, and Phillip Isola. The platonic representation hypothesis, 2024. URL https://arxiv.org/abs/2405.07987. ↗
  13. 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. arXiv preprint arXiv:2412.16720, 2024.
  14. Ai Jian, Weijie Qiu, Xiaokun Wang, Peiyu Wang, Yunzhuo Hao, Jiangbo Pei, Yichen Wei, Yi Peng, and Xuchen Song. Csvqa: A chinese multimodal benchmark for evaluating stem reasoning capabilities of vlms, 2025. URL https://arxiv.org/abs/2505.24120. — [arXiv:2505.24120](https://arxiv.org/abs/2505.24120) ↗
  15. Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning, 2023. URL https://arxiv.org/abs/2304.08485. ↗
  16. Yuan Liu, Haodong Duan, Yuanhan Zhang, Bo Li, Songyang Zhang, Wangbo Zhao, Yike Yuan, Jiaqi Wang, Conghui He, Ziwei Liu, et al. Mmbench: Is your multi-modal model an all-around player? In European conference on computer vision, pp. 216–233. Springer, 2024.
  17. Pan Lu, Hritik Bansal, Tony Xia, Jiacheng Liu, Chunyuan Li, Hannaneh Hajishirzi, Hao Cheng, Kai-Wei Chang, Michel Galley, and Jianfeng Gao. Mathvista: Evaluating mathematical reasoning of foundation models in visual contexts. arXiv preprint arXiv:2310.02255, 2023.
  18. Fanqing Meng, Lingxiao Du, Zongkai Liu, Zhixiang Zhou, Quanfeng Lu, Daocheng Fu, Botian Shi, Wenhai Wang, Junjun He, Kaipeng Zhang, et al. Mm-eureka: Exploring visual aha moment with rule-based large-scale reinforcement learning. arXiv preprint arXiv:2503.07365, 2025. find
  19. Youssef Mroueh. Reinforcement learning with verifiable rewards: Grpo’s effective loss, dynamics, and success amplification, 2025. URL https://arxiv.org/abs/2503.06639. — [arXiv:2503.06639](https://arxiv.org/abs/2503.06639) ↗
  20. Dang Nguyen, Jian Chen, Yu Wang, et al. Gui agents: A survey, 2024. URL https://arxiv.org/abs/24 12.13501. ↗
  21. OpenAI. Gpt-4o system card, 2024. URL https://openai.com/index/hello-gpt-4o/. ↗
  22. Yi Peng, Peiyu Wang, 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. Skywork r1v: Pioneering multimodal reasoning with chain-of-thought, 2025. URL https://arxiv.org/abs/2504.0 5599. ↗
  23. Runqi Qiao, Qiuna Tan, Guanting Dong, Minhui Wu, Chong Sun, Xiaoshuai Song, Zhuoma GongQue, Shanglin Lei, Zhe Wei, Miaoxuan Zhang, Runfeng Qiao, Yifan Zhang, Xiao Zong, Yida Xu, Muxi Diao, Zhimin Bao, Chen Li, and Honggang Zhang. We-math: Does your large multimodal model achieve human-like mathemati… find
  24. John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms, 2017. URL https://arxiv.org/abs/1707.06347. ↗
  25. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, Y. K. Li, Y. Wu, and Daya Guo. Deepseekmath: Pushing the limits of mathematical reasoning in open language models, 2024. URL https://arxiv.org/abs/2402.03300. ↗
  26. Hui Shen, Taiqiang Wu, Qi Han, Yunta Hsieh, Jizhou Wang, Yuyue Zhang, Yuxin Cheng, Zijian Hao, Yuansheng Ni, Xin Wang, Zhongwei Wan, Kai Zhang, Wendong Xu, Jing Xiong, Ping Luo, Wenhu Chen, Chaofan Tao, Zhuoqing Mao, and Ngai Wong. Phyx: Does your model have the ”wits” for physical reasoning?, 2025.… — [arXiv:2505.15929](https://arxiv.org/abs/2505.15929) find
  27. Guangming Sheng, Chi Zhang, Zilingfeng Ye, Xibin Wu, Wang Zhang, Ru Zhang, Yanghua Peng, Haibin Lin, and Chuan Wu. Hybridflow: A flexible and efficient rlhf framework. arXiv preprint arXiv: 2409.19256, 2024.
  28. Core Team, Zihao Yue, Zhenru Lin, Yifan Song, Weikun Wang, et al. Mimo-vl technical report, 2025. URL https://arxiv.org/abs/2506.03569. ↗
  29. Qwen Team. Qvq: To see the world with wisdom. https://qwenlm.github.io/blog/qvq-72b-previ ew/, 2024. ↗
  30. Qwen Team. Qwq-32b: Embracing the power of reinforcement learning, March 2025. URL https://qwenlm.github.io/blog/qwq-32b/. ↗
  31. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. Llama: Open and efficient foundation language models, 2023. URL ht…
  32. Beichen Wang, Juexiao Zhang, Shuwen Dong, Irving Fang, and Chen Feng. Vlm see, robot do: Human demo video to robot action plan via vision language model, 2024a. URL https://arxiv.org/abs/24 10.08792. ↗
  33. Chenlong Wang, Yuanning Feng, Dongping Chen, Zhaoyang Chu, Ranjay Krishna, and Tianyi Zhou. Wait, we don’t need to ”wait”! removing thinking tokens improves reasoning efficiency, 2025a. URL https://arxiv.org/abs/2506.08343. ↗
  34. Peiyu Wang, Yichen Wei, Yi Peng, Xiaokun Wang, Weijie Qiu, Wei Shen, Tianyidan Xie, Jiangbo Pei, Jianhao Zhang, Yunzhuo Hao, Xuchen Song, Yang Liu, and Yahui Zhou. Skywork r1v2: Multimodal hybrid reinforcement learning for reasoning, 2025b. URL https://arxiv.org/abs/2504.16656. ↗
  35. Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Yang Fan, Kai Dang, Mengfei Du, Xuancheng Ren, Rui Men, Dayiheng Liu, Chang Zhou, Jingren Zhou, and Junyang Lin. Qwen2-vl: Enhancing vision-language model’s perception of the world a… — Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution
  36. Shenzhi Wang, Le Yu, Chang Gao, Chujie Zheng, Shixuan Liu, Rui Lu, Kai Dang, Xionghui Chen, Jianxin Yang, Zhenru Zhang, Yuqiong Liu, An Yang, Andrew Zhao, Yang Yue, Shiji Song, Bowen Yu, Gao Huang, and Junyang Lin. Beyond the 80/20 rule: High-entropy minority tokens drive effective reinforcement lea… — Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning
  37. Xiaokun Wang, Peiyu Wang, Jiangbo Pei, Wei Shen, Yi Peng, Yunzhuo Hao, Weijie Qiu, Ai Jian, Tianyidan Xie, Xuchen Song, Yang Liu, and Yahui Zhou. Skywork-vl reward: An effective reward model for multimodal understanding and reasoning, 2025d. URL https://arxiv.org/abs/2505.07263. — [arXiv:2505.07263](https://arxiv.org/abs/2505.07263) ↗
  38. 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. Advances in neural information processing systems, 35:24824–24837, 2022.
  39. Kun Xiang, Heng Li, Terry Jingchen Zhang, Yinya Huang, Zirong Liu, Peixin Qu, Jixi He, Jiaqi Chen, Yu-Jie Yuan, Jianhua Han, Hang Xu, Hanhui Li, Mrinmaya Sachan, and Xiaodan Liang. Seephys: Does seeing help thinking? – benchmarking vision-based physics reasoning, 2025. URL https://arxiv.org/abs/250… ↗
  40. Yijia Xiao, Edward Sun, Tianyu Liu, and Wei Wang. Logicvista: Multimodal llm logical reasoning benchmark in visual contexts, 2024. URL https://arxiv.org/abs/2407.04973. — [arXiv:2407.04973](https://arxiv.org/abs/2407.04973) ↗
  41. Weiye Xu, Jiahao Wang, Weiyun Wang, Zhe Chen, Wengang Zhou, Aijun Yang, Lewei Lu, Houqiang Li, Xiaohua Wang, Xizhou Zhu, Wenhai Wang, Jifeng Dai, and Jinguo Zhu. Visulogic: A benchmark for evaluating visual reasoning in multi-modal large language models, 2025. URL https://arxiv.org/abs/2504.15279. — [arXiv:2504.15279](https://arxiv.org/abs/2504.15279) ↗
  42. Yi Xu, Yuxin Hu, Zaiwei Zhang, Gregory P. Meyer, Siva Karthik Mustikovela, Siddhartha Srinivasa, Eric M. Wolff, and Xin Huang. Vlm-ad: End-to-end autonomous driving through vision-language model supervision, 2024. URL https://arxiv.org/abs/2412.14446. — [arXiv:2412.14446](https://arxiv.org/abs/2412.14446) ↗
  43. Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Weinan Dai, Tiantian Fan, Gaohong Liu, Lingjun Liu, Xin Liu, Haibin Lin, Zhiqi Lin, Bole Ma, Guangming Sheng, Yuxuan Tong, Chi Zhang, Mofan Zhang, Wang Zhang, Hang Zhu, Jinhua Zhu, Jiaze Chen, Jiangjie Chen, Chengyi Wang, Hongli… — DAPO: An Open-Source LLM Reinforcement Learning System at Scale
  44. Jiakang Yuan, Tianshuo Peng, Yilei Jiang, Yiting Lu, Renrui Zhang, Kaituo Feng, Chaoyou Fu, Tao Chen, Lei Bai, Bo Zhang, and Xiangyu Yue. Mme-reasoning: A comprehensive benchmark for logical reasoning in mllms, 2025. URL https://arxiv.org/abs/2505.21327. — [arXiv:2505.21327](https://arxiv.org/abs/2505.21327) ↗
  45. Xiang Yue, Yuansheng Ni, Kai Zhang, Tianyu Zheng, Ruoqi Liu, Ge Zhang, Samuel Stevens, Dongfu Jiang, Weiming Ren, Yuxuan Sun, et al. Mmmu: A massive multi-discipline multimodal understanding and reasoning benchmark for expert agi. In Proceedings of the IEEE/CVF Conference on Computer Vision and Patt…
  46. 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. Mmmu-pro: A more robust multi-discipline multimodal understanding benchmark, 2025. URL https://arxiv.org/abs/2409.02813. ↗
  47. Jihai Zhang, Tianle Li, Linjie Li, Zhengyuan Yang, and Yu Cheng. Are unified vision-language models necessary: Generalization across understanding and generation, 2025. URL https://arxiv.org/abs/2505.23043. — [arXiv:2505.23043](https://arxiv.org/abs/2505.23043) ↗
  48. Renrui Zhang, Dongzhi Jiang, Yichi Zhang, Haokun Lin, Ziyu Guo, Pengshuo Qiu, Aojun Zhou, Pan Lu, Kai-Wei Chang, Peng Gao, and Hongsheng Li. Mathverse: Does your multi-modal llm truly see the diagrams in visual math problems?, 2024. URL https://arxiv.org/abs/2403.14624. ↗
  49. Jun Zhou, Jiahao Li, Zunnan Xu, Hanhui Li, Yiji Cheng, Fa-Ting Hong, Qin Lin, Qinglin Lu, and Xiaodan Liang. Fireedit: Fine-grained instruction-based image editing via region-aware vision language model, 2025. URL https://arxiv.org/abs/2503.19839. — [arXiv:2503.19839](https://arxiv.org/abs/2503.19839) ↗
  50. Jinguo Zhu, Weiyun Wang, Zhe Chen, Zhaoyang Liu, Shenglong Ye, Lixin Gu, Hao Tian, Yuchen Duan, Weijie Su, Jie Shao, Zhangwei Gao, Erfei Cui, Xuehui Wang, Yue Cao, Yangzhou Liu, Xingguang Wei, Hongjie Zhang, Haomin Wang, Weiye Xu, Hao Li, Jiahao Wang, Nianchen Deng, Songze Li, Yinan He, Tan Jiang, J… — InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/shen-2025-skywork-r1v3-technical-report/