Paper Detail
Kwai Keye Team, Bin Wen, Changyi Liu, Chengru Song, Chongling Rao, Guowang Zhang, Han Li, Haonan Fan, Hengrui Ju, Jiankang Chen, Jiapeng Chen, Jiawei Yuan, Kaixuan Yang, Kaiyu Jiang, Kun Gai, Lingzhi Zhou, Na Nie, Sen Na, Tianke Zhang, Tingting Gao, Xuanyu Zheng, Yulong Chen, Fan Yang, Haixuan Gao, Lele Yang, Mingqiao Liu, Muxi Diao, Qi Zhang, Qile Su, Wei Chen, Wentao Hong, Xingyu Lu, Yancheng Long, Yankai Yang, Yingxin Li, Yiyang Fan, Yu Xia, Yuzhe Chen, Ziliang Lai, Chuan Yi, Haonan Jia, Tianming Liang, Weixin Xu, Xiaoxiao Ma, Yang Tian, Yufei Han, Feng Han, Hang Li, Jing Wang, Jinghui Jia, Junmin Chen, Junyu Shi, Ruilin Zhang
We introduce Kwai Keye-VL-2.0-30B-A3B, an open-source Mixture-of-Experts (MoE) multimodal foundation model designed to advance long-video understanding and agentic intelligence. To address the challenges of ultra-long contexts, information redundancy, and prohibitive computational costs inherent in hour-level videos, Keye-VL-2.0 is the first to adapt DeepSeek Sparse Attention (DSA) to GQA-based multimodal architectures, enabling lossless 256K context processing while capturing critical frames and long-range temporal dependencies. This architecture is underpinned by a highly optimized training and inference infrastructure, including scalable video I/O, heterogeneous ViT-LM parallelism, and custom DSA kernels that significantly maximize throughput and minimize computational overhead. Furthermore, to overcome the algorithmic dilemma of catastrophic forgetting during multi-task alignment, we introduce Cross-Modal Multi-Teacher On-Policy Distillation (MOPD) paired with Context-RL and Video-RL. By distilling dense token-level teacher feedback from on-policy rollouts back into the MoE backbone, which activates only 3B parameters, Keye-VL-2.0 natively empowers advanced agent collaboration across Code, Tool, and Search scenarios with multimodal self-correction. Extensive evaluations across video understanding, temporal grounding, reasoning, STEM, and agent benchmarks demonstrate that Keye-VL-2.0-30B-A3B achieves state-of-the-art performance among models of similar scale, particularly excelling in fine-grained temporal localization on TimeLens and long-video comprehension on Video-MME-v2 and LongVideoBench. We release our model checkpoints to accelerate community progress toward scalable and robust multimodal agentic applications.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@misc{team2026kwai,
title = {Kwai Keye-VL-2.0 Technical Report},
author = {Kwai Keye Team and Bin Wen and Changyi Liu and Chengru Song and Chongling Rao and Guowang Zhang and Han Li and Haonan Fan and Hengrui Ju and Jiankang Chen and Jiapeng Chen and Jiawei Yuan and Kaixuan Yang and Kaiyu Jiang and Kun Gai and Lingzhi Zhou and Na Nie and Sen Na and Tianke Zhang and Tingting Gao and Xuanyu Zheng and Yulong Chen and Fan Yang and Haixuan Gao and Lele Yang and Mingqiao Liu and Muxi Diao and Qi Zhang and Qile Su and Wei Chen and Wentao Hong and Xingyu Lu and Yancheng Long and Yankai Yang and Yingxin Li and Yiyang Fan and Yu Xia and Yuzhe Chen and Ziliang Lai and Chuan Yi and Haonan Jia and Tianming Liang and Weixin Xu and Xiaoxiao Ma and Yang Tian and Yufei Han and Feng Han and Hang Li and Jing Wang and Jinghui Jia and Junmin Chen and Junyu Shi and Ruilin Zhang},
year = {2026},
abstract = {We introduce Kwai Keye-VL-2.0-30B-A3B, an open-source Mixture-of-Experts (MoE) multimodal foundation model designed to advance long-video understanding and agentic intelligence. To address the challenges of ultra-long contexts, information redundancy, and prohibitive computational costs inherent in hour-level videos, Keye-VL-2.0 is the first to adapt DeepSeek Sparse Attention (DSA) to GQA-based multimodal architectures, enabling lossless 256K context processing while capturing critical frames an},
url = {https://huggingface.co/papers/2606.10651},
keywords = {Mixture-of-Experts, multimodal foundation model, DeepSeek Sparse Attention, GQA-based architectures, 256K context processing, heterogeneous ViT-LM parallelism, custom DSA kernels, Cross-Modal Multi-Teacher On-Policy Distillation, Context-RL, Video-RL, dense token-level teacher feedback, on-policy rollouts, agent collaboration, Code, Tool, Search scenarios, multimodal self-correction, code available, huggingface daily},
eprint = {2606.10651},
archiveprefix = {arXiv},
}
{}