Kwai Keye-VL-2.0 Technical Report

Hugging Face Daily Papers Papers
mixture-of-experts multimodal long-video-understanding deepseek-sparse-attention distillation reinforcement-learning open-source

Summary

This technical report presents Kwai Keye-VL-2.0, an open-source Mixture-of-Experts multimodal foundation model designed for long-video understanding and agentic intelligence, leveraging DeepSeek Sparse Attention and cross-modal distillation to achieve state-of-the-art performance among similar-scale models.

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.
Original Article
View Cached Full Text

Cached at: 06/10/26, 09:43 AM

Paper page - Kwai Keye-VL-2.0 Technical Report

Source: https://huggingface.co/papers/2606.10651 Published on Jun 9

#2 Paper of the day Authors:

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

Abstract

Kwai Keye-VL-2.0-30B-A3B is an open-source Mixture-of-Experts multimodal foundation model that enables long-video understanding and agentic intelligence through DeepSeek Sparse Attention and specialized training infrastructure.

We introduce Kwai Keye-VL-2.0-30B-A3B, an open-sourceMixture-of-Experts(MoE)multimodal foundation modeldesigned 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 adaptDeepSeek Sparse Attention(DSA) to GQA-based multimodal architectures, enabling lossless256K context processingwhile 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, andcustom DSA kernelsthat significantly maximize throughput and minimize computational overhead. Furthermore, to overcome the algorithmic dilemma of catastrophic forgetting during multi-task alignment, we introduceCross-Modal Multi-Teacher On-Policy Distillation(MOPD) paired withContext-RLandVideo-RL. By distillingdense token-level teacher feedbackfromon-policy rolloutsback into the MoE backbone, which activates only 3B parameters, Keye-VL-2.0 natively empowers advancedagent collaborationacrossCode,Tool, andSearch scenarioswithmultimodal 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.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2606\.10651

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper1

#### Kwai-Keye/Keye-VL-2.0-30B-A3B Image-Text-to-Text• 31B• Updatedabout 5 hours ago • 3.25k • 109

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.10651 in a dataset README.md to link it from this page.

Spaces citing this paper1

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Kwai-Keye/Keye-VL-2.0-30B-A3B

Hugging Face Models Trending

Kwai-Keye releases Keye-VL-2.0-30B-A3B, a 30B-class vision-language model with advanced video understanding, sparse attention, and agent capabilities, achieving top benchmarks.

Kimi K3: Open Frontier Intelligence

Hugging Face Daily Papers

Kimi K3 is a 2.8 trillion parameter Mixture-of-Experts model with 104 billion active parameters, native vision, and a 1-million-token context window, achieving frontier-level performance across multiple domains and released as open weights.

VideoChat3: Fully Open Video MLLM for Efficient and Generalist Video Understanding

Papers with Code Trending

VideoChat3 is a fully open, efficient, and generalist video-centric multimodal large language model that introduces Inflated 3D Vision Transformer (I3D-ViT) and Adaptive Frame Resolution for streaming video perception, along with scalable video data synthesis pipelines, achieving superior performance with only 4B parameters.