---
title: "Kwai Keye-VL-2.0 Technical Report"
canonical_url: "https://www.modelscope.cn/papers/294691"
md_url: "https://www.modelscope.cn/papers/294691.md"
arxiv_id: 2606.10651
published: 2026-06-09
last_updated: 2026-06-09
authors:
  - "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"
model_name: "Kwai Keye-VL-2.0-30B-A3B"
model_developer: "快手集团（Kuaishou Group）"
domain:
  - "计算机视觉"
  - "多模态大模型"
  - "长视频理解"
  - "智能体"
  - "稀疏注意力"
type:
  - "计算机视觉"
  - "多模态大模型"
  - "长视频理解"
  - "智能体"
  - "稀疏注意力"
  - cs.CV
arxiv_url: "https://arxiv.org/abs/2606.10651"
pdf_url: "https://arxiv.org/pdf/2606.10651"
---

# Kwai Keye-VL-2.0 Technical Report

> 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…

「Kwai Keye-VL-2.0 Technical Report」是 ModelScope 魔搭社区收录的论文，arXiv 2606.10651，作者为 Kwai Keye Team, Bin Wen, Changyi Liu et al.，发表于 2026-06-09，属于 计算机视觉、多模态大模型、长视频理解 领域。

- **ArXiv**: 2606.10651
- **Published**: 2026-06-09
- **Authors**: 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
- **Model**: Kwai Keye-VL-2.0-30B-A3B
- **Developer**: 快手集团（Kuaishou Group）
- **Domain**: 计算机视觉, 多模态大模型, 长视频理解, 智能体, 稀疏注意力
- **ArXiv URL**: https://arxiv.org/abs/2606.10651
- **PDF**: https://arxiv.org/pdf/2606.10651

Source: https://www.modelscope.cn/papers/294691

---

## 摘要

Kwai Keye-VL-2.0-30B-A3B 是由快手团队开发的开源混合专家（MoE）多模态基础模型，专为长视频理解和智能体（Agent）任务设计。该模型总参数量为300亿，激活参数为30亿，首次将DeepSeek稀疏注意力（DSA）机制引入基于GQA的多模态架构，实现了256K超长上下文的无损处理。此外，模型采用跨模态多教师同策略蒸馏（MOPD）以及多种强化学习（Context-RL、Video-RL）策略，有效缓解了多任务对齐中的灾难性遗忘与视觉幻觉问题，在长视频理解、时序定位及代码与工具调用等智能体基准测试中达到了同规模模型的领先性能。

## 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 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.