Year: 2026
Type: preprint
Abstract: We present MOSS-VL, an open vision-language model family that treats real-time interaction -- perceiving while it speaks -- as a first-class capability. It is co-designed across the stack: the language decoder attends to vision only through gated cross-attention, so the model can naturally see incoming frames while generating; a synthesized interaction corpus supervises when to speak, when to stay silent, and when to revise; and a staged curriculum concentrates all real-time-specific training in one light final stage over a strong offline foundation. Offline, MOSS-VL-Instruct is competitive at comparable scale and leads temporal-reasoning video sets. Across four streaming benchmarks, MOSS-VL-Realtime posts the best average on three (second on the fourth) among open-source streaming models, sweeping the three subsets that squarely test proactive behavior -- 66.0 vs. 37.5 for the best baseline on OmniMMI Proactive Alerting. With 11.3B parameters but visual tokens outside the decoded sequence, MOSS-VL widens its time-to-first-token advantage over same-backbone Qwen3-VL-8B from 2.8x to 5.1x as visual context grows. We release all five checkpoints, the training curriculum, and the real-time inference code at https://github.com/OpenMOSS/MOSS-VL.
Source: arXiv (Cornell University)
Authors: Wang, Pengyu, Tan, Chenkun, Zhou, Shaojun, Zhou, Qirui, Chen, Yanxin, He, Xingyang, Zeng, Huazheng, Cheng, Jijun, Wang, Chenghao, Qian, Xiaomeng, Wang, Pengfei, Huang, Zhan, Gao, Shanqing, Huang, Wei, Cao, Longjun, Ran, Wu, Liu, Jie, Zhu, Changtai, Wang, Hongkai, Tian, Yixian, Liu, Chenghao, Ye, Zhen, Wang, Xinghao, Jiang, Botian, Feng, Guoguo, Fei, Zhaoye, Li, Ruixiao, Chen, Mingshu, Gao, Yang, Cheng, Qinyuan, Li, Shimin, Qiu, Xipeng
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7203670747
Cites:
Cited by:
Related to: