美團 LongCat 整合 WBench 評測模型權重
meituan-longcat/WBench-weights
美團 LongCat 的 meituan-longcat/WBench-weights 儲存庫整合 WBench 評測模型權重,供互動影片世界模型評估使用。WBench 從影片質素、設定遵循、互動遵循、一致性及物理規則符合度五個維度評估,涵蓋 289 個測試案例及 1,058 個互動回合。這些權重僅供學術研究及評估用途。

Pre-trained model weights for WBench evaluation.
This repository contains the consolidated model weights for WBench, a comprehensive multi-turn benchmark for interactive video world model evaluation. WBench evaluates world models along five dimensions: video quality, setting adherence, interaction adherence, consistency, and physics compliance. It contains 289 test cases and 1,058 interaction turns covering diverse scenes, styles, subjects, and perspectives.
Usage
Please refer to the WBench GitHub repository for installation and evaluation instructions. You can download the weights using the Hugging Face CLI:
huggingface-cli download meituan-longcat/WBench-weights --local-dir weights/
Disclaimer
We consolidate these weights into a single repository to help the community quickly deploy the WBench evaluation framework without hunting for individual checkpoints. These weights are redistributed solely for academic research and evaluation purposes. All rights belong to the original authors. See LICENSE_NOTICE.md for per-model licenses. If you believe any content infringes your rights, please contact us and we will remove it promptly:
- Kaining Ying: kaining.ying.cv@gmail.com
- Siyu Ren: rensiyu07@meituan.com
Citation
@article{ying2026wbenchcomprehensivemultiturnbenchmark,
title={WBench: A Comprehensive Multi-turn Benchmark for Interactive Video World Model Evaluation},
author={Ying, Kaining and Hu, Hengrui and Ren, Siyu and Li, Jiamu and Chen, Fengjiao and Wang, Ziwen and Cao, Xuezhi and Cai, Xunliang and Ding, Henghui},
journal={arXiv preprint arXiv:2605.25874},
year={2026}
}
來源:美團 LongCat:HuggingFace 新模型 · huggingface.co