GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning

Wenyi Hong, Wenmeng Yu, Xiaotao Gu, Guo Wang, GuobingGan, Haomiao Tang, Jiale Cheng, Ji Qi, Junhui Ji, kinnplh, ShuaiqiDuan, 王维汉, Yan, Yean Cheng, Zehai He, Zhe Su, ZhenYang21, Ziyang Pan, Zeng Aohan, wangbaoxu, Boyan Shi, Pcy, Chenhui Zhang, Da Yin, Fan Yang, Guoqing Chen, Jiazheng Xu, Gary Chen, chenjing, Jinhao Chen, linlincode, Jinjiang Wang, Chen Junjie, LEI Le-qi, Leyi Pan, Mingzhi Zhang, Qinkai Zheng, Sheng Yang, Shi Zhong, Shiyu Huang, ZhaoShuyuan, Sean Xue, Shangqing Tu, mengshengbiao, Tianshu Zhang, 罗天蔚, Tianxiang Hao, Tianle Gong, liwenkai, Wei Jia, Xin Lv, Huang Xuancheng, Yanling Wang, Xue Yadong, Yanfeng Wang, A1phaN, Evan Du, Yiming Shi, yiheng huang, Yilin Niu, Yuan Wang, Yuanchang Yue, Yuchen Li, zyt, zR, Zhengxiao Du, Zhenyu Hou, xuezhao, Zhengxiao Du, Zihan Wang, Peng Zhang, Debing Liu, Bin Xu, Juanzi Li, Minlie Huang, Yuxiao Dong, Jie Tang

GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning: 257 upvotes on Hugging Face Daily Papers, #1 of 18 papers on 2025-07-02. Day-by-day upvote history.

We present GLM-4.1V-Thinking, a vision-language model (VLM) designed to advance general-purpose multimodal reasoning. In this report, we share our key findings in the development of the reasoning-centric training framework. We first develop a capable vision foundation model with significant potential through large-scale pre-training, which arguably sets the upper bound for the final performance. Reinforcement Learning with Curriculum Sampling (RLCS) then unlocks the full potential of the model, leading to comprehensive capability enhancement across a diverse range of tasks, including STEM problem solving, video understanding, content recognition, coding, grounding, GUI-based agents, and long document understanding, among others. To facilitate research in this field, we open-source GLM-4.1V-9B-Thinking, which achieves state-of-the-art performance among models of comparable size. In a comprehensive evaluation across 28 public benchmarks, our model outperforms Qwen2.5-VL-7B on nearly all tasks and achieves comparable or even superior performance on 18 benchmarks relative to the significantly larger Qwen2.5-VL-72B. Notably, GLM-4.1V-9B-Thinking also demonstrates competitive or superior performance compared to closed-source models such as GPT-4o on challenging tasks including long document understanding and STEM reasoning, further underscoring its strong capabilities. Code, models and more information are released at https://github.com/THUDM/GLM-4.1V-Thinking.

Paper page on Hugging Face · arXiv

Data: hysts-bot-data/daily-papers-stats and the Daily Papers API. Open data: tardellirs/paper-pulse-data. Sister project: Model Pulse, the download history of every model on the Hub.