MiMo-VL Technical Report
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866915324166668288 |
|---|---|
| author | Core Team Yue, Zihao Lin, Zhenru Song, Yifan Wang, Weikun Ren, Shuhuai Gu, Shuhao Li, Shicheng Li, Peidian Zhao, Liang Li, Lei Bao, Kainan Tian, Hao Zhang, Hailin Wang, Gang Zhu, Dawei Cici He, Chenhong Ye, Bowen Shen, Bowen Zhang, Zihan Jiang, Zihan Zheng, Zhixian Song, Zhichao Luo, Zhenbo Yu, Yue Wang, Yudong Tian, Yuanyuan Tu, Yu Yan, Yihan Huang, Yi Wang, Xu Xu, Xinzhe Song, Xingchen Zhang, Xing Yong, Xing Zhang, Xin Deng, Xiangwei Yang, Wenyu Ma, Wenhan Lv, Weiwei Zhuang, Weiji Liu, Wei Deng, Sirui Liu, Shuo Chen, Shimao Yu, Shihua Liu, Shaohui Wang, Shande Ma, Rui Wang, Qiantong Wang, Peng Chen, Nuo Zhu, Menghang Zhou, Kangyang Zhou, Kang Fang, Kai Shi, Jun Dong, Jinhao Xiao, Jiebao Xu, Jiaming Liu, Huaqiu Xu, Hongshen Qu, Heng Zhao, Haochen Lv, Hanglong Wang, Guoan Zhang, Duo Zhang, Dong Zhang, Di Ma, Chong Liu, Chang Cai, Can Xia, Bingquan |
| author_facet | Core Team Yue, Zihao Lin, Zhenru Song, Yifan Wang, Weikun Ren, Shuhuai Gu, Shuhao Li, Shicheng Li, Peidian Zhao, Liang Li, Lei Bao, Kainan Tian, Hao Zhang, Hailin Wang, Gang Zhu, Dawei Cici He, Chenhong Ye, Bowen Shen, Bowen Zhang, Zihan Jiang, Zihan Zheng, Zhixian Song, Zhichao Luo, Zhenbo Yu, Yue Wang, Yudong Tian, Yuanyuan Tu, Yu Yan, Yihan Huang, Yi Wang, Xu Xu, Xinzhe Song, Xingchen Zhang, Xing Yong, Xing Zhang, Xin Deng, Xiangwei Yang, Wenyu Ma, Wenhan Lv, Weiwei Zhuang, Weiji Liu, Wei Deng, Sirui Liu, Shuo Chen, Shimao Yu, Shihua Liu, Shaohui Wang, Shande Ma, Rui Wang, Qiantong Wang, Peng Chen, Nuo Zhu, Menghang Zhou, Kangyang Zhou, Kang Fang, Kai Shi, Jun Dong, Jinhao Xiao, Jiebao Xu, Jiaming Liu, Huaqiu Xu, Hongshen Qu, Heng Zhao, Haochen Lv, Hanglong Wang, Guoan Zhang, Duo Zhang, Dong Zhang, Di Ma, Chong Liu, Chang Cai, Can Xia, Bingquan |
| contents | We open-source MiMo-VL-7B-SFT and MiMo-VL-7B-RL, two powerful vision-language models delivering state-of-the-art performance in both general visual understanding and multimodal reasoning. MiMo-VL-7B-RL outperforms Qwen2.5-VL-7B on 35 out of 40 evaluated tasks, and scores 59.4 on OlympiadBench, surpassing models with up to 78B parameters. For GUI grounding applications, it sets a new standard with 56.1 on OSWorld-G, even outperforming specialized models such as UI-TARS. Our training combines four-stage pre-training (2.4 trillion tokens) with Mixed On-policy Reinforcement Learning (MORL) integrating diverse reward signals. We identify the importance of incorporating high-quality reasoning data with long Chain-of-Thought into pre-training stages, and the benefits of mixed RL despite challenges in simultaneous multi-domain optimization. We also contribute a comprehensive evaluation suite covering 50+ tasks to promote reproducibility and advance the field. The model checkpoints and full evaluation suite are available at https://github.com/XiaomiMiMo/MiMo-VL. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_03569 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | MiMo-VL Technical Report Core Team Yue, Zihao Lin, Zhenru Song, Yifan Wang, Weikun Ren, Shuhuai Gu, Shuhao Li, Shicheng Li, Peidian Zhao, Liang Li, Lei Bao, Kainan Tian, Hao Zhang, Hailin Wang, Gang Zhu, Dawei Cici He, Chenhong Ye, Bowen Shen, Bowen Zhang, Zihan Jiang, Zihan Zheng, Zhixian Song, Zhichao Luo, Zhenbo Yu, Yue Wang, Yudong Tian, Yuanyuan Tu, Yu Yan, Yihan Huang, Yi Wang, Xu Xu, Xinzhe Song, Xingchen Zhang, Xing Yong, Xing Zhang, Xin Deng, Xiangwei Yang, Wenyu Ma, Wenhan Lv, Weiwei Zhuang, Weiji Liu, Wei Deng, Sirui Liu, Shuo Chen, Shimao Yu, Shihua Liu, Shaohui Wang, Shande Ma, Rui Wang, Qiantong Wang, Peng Chen, Nuo Zhu, Menghang Zhou, Kangyang Zhou, Kang Fang, Kai Shi, Jun Dong, Jinhao Xiao, Jiebao Xu, Jiaming Liu, Huaqiu Xu, Hongshen Qu, Heng Zhao, Haochen Lv, Hanglong Wang, Guoan Zhang, Duo Zhang, Dong Zhang, Di Ma, Chong Liu, Chang Cai, Can Xia, Bingquan Computation and Language We open-source MiMo-VL-7B-SFT and MiMo-VL-7B-RL, two powerful vision-language models delivering state-of-the-art performance in both general visual understanding and multimodal reasoning. MiMo-VL-7B-RL outperforms Qwen2.5-VL-7B on 35 out of 40 evaluated tasks, and scores 59.4 on OlympiadBench, surpassing models with up to 78B parameters. For GUI grounding applications, it sets a new standard with 56.1 on OSWorld-G, even outperforming specialized models such as UI-TARS. Our training combines four-stage pre-training (2.4 trillion tokens) with Mixed On-policy Reinforcement Learning (MORL) integrating diverse reward signals. We identify the importance of incorporating high-quality reasoning data with long Chain-of-Thought into pre-training stages, and the benefits of mixed RL despite challenges in simultaneous multi-domain optimization. We also contribute a comprehensive evaluation suite covering 50+ tasks to promote reproducibility and advance the field. The model checkpoints and full evaluation suite are available at https://github.com/XiaomiMiMo/MiMo-VL. |
| title | MiMo-VL Technical Report |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2506.03569 |