Kwai Keye-VL

model paper
Kuaishou
8B parameter multimodal foundation model specializing in short-video understanding, with accompanying technical report.
GitHubHuggingFaceAnnouncement

Outputs 3

Kwai Keye-VL

model

8B parameter multimodal foundation model specializing in short-video understanding.

GitHubHuggingFaceAnnouncement
Architecture DENSE
Parameters 8B

Kwai Keye-VL Technical Report

paper

Technical report for the Kwai Keye-VL multimodal foundation model.

GitHubHuggingFace

Keye-VL-2.0-30B-A3B

model

<p>Major flagship upgrade announced May 25, weights uploaded May 29. 31B MoE multimodal with <strong>DeepSeek Sparse Attention</strong> (DSA), 256K context, and a built-in Search / Tool / Code agent. First production multimodal model using DSA. <strong>LongVideoBench 74.1</strong> &mdash; beats Qwen3.5-35B and Qwen3-VL-235B on long-video temporal benchmarks.</p>

HuggingFaceGitHub
Architecture MOE
Parameters 31B
Active params 3B
Context window 262,144
multimodalvideoopen-weight