WeMM-Video open-source omni-modal model family released
The authors released WeMM-Video, a Chinese-English bilingual family of omni-modal foundation models in 2B, 4B, 9B, and 35B-A3B sizes. The models jointly process text, images, audio, and video, offering non-thinking and thinking modes, and were trained on an approximately 1.2-trillion-token multimodal corpus. The paper reports evaluation across more than 120 benchmarks and says the authors also released the model family, evaluation pipeline, and reproduction pipeline.
WeMM-Video adds four openly released model sizes with unified audio-visual reasoning and bilingual support, while its reported state-of-the-art results remain author claims rather than independently confirmed findings.
Computer Science > Computer Vision and Pattern Recognition
arXiv:2611.26409 (cs)
[Submitted on 19 Nov 2026]
Title:WeMM-Video: Pushing the Frontiers of Open-Source Omni-Modal Video Understanding
Authors:Changli Tang, Jie Yang, Tianyi Wang, Lei Li, Daoze Zhang, Zijie Xin, Zhan Liu, Li Li, Jinchang Xu, Fengyun Rao, Chao Zhang, Jing Lyu
View a PDF of the paper titled WeMM-Video: Pushing the Frontiers of Open-Source Omni-Modal Video Understanding, by Changli Tang and 11 other authors
#WeMM-Video#WeMM-Video-2B#WeMM-Video-4B#WeMM-Video-9B#WeMM-Video-35B-A3B#Qwen3.5-Omni-Flash#Seed2.0-Mini-0428#Gemini-3.0-Flash#Qwen3-Omni-30B-A3B-Instruct