Qwen3-VL-32B-Thinking-Desktop-Teacher

Qwen3-VL-32B-Thinking-Desktop-Teacher

This is the Desktop Teacher model from the UI-MOPD project — a platform-specific expert trained for desktop GUI interaction tasks.

Model Description

Qwen3-VL-32B-Thinking-Desktop-Teacher is fine-tuned from Qwen3-VL-32B-Thinking on desktop interaction trajectories from the Uni-GUI dataset. It serves as the desktop-platform teacher in the UI-MOPD multi-teacher on-policy distillation framework.

Key Highlights

• Base Model: Qwen3-VL-32B-Thinking
• Training Data: Desktop subset of Uni-GUI (~160K interaction steps across ~11.5K trajectories)
• Role: Platform-specific teacher for desktop environments in the UI-MOPD distillation pipeline
• OSWorld Performance: 46.3% task success rate (vs. 41.0% base model)

Training Details

This model is obtained in Stage 1 of the UI-MOPD training pipeline:

1. Stage 1 (This Model): Supervised fine-tuning of Qwen3-VL-32B-Thinking on desktop GUI interaction trajectories from Uni-GUI to produce a platform-specific desktop expert.
2. Stage 2: The desktop teacher (this model) and a mobile teacher jointly guide a shared 8B student policy via multi-teacher on-policy distillation with platform-conditioned routing.

Performance

Intended Use

This model is designed to:

• Serve as a teacher model in the UI-MOPD distillation framework for training cross-platform GUI agents
• Be used as a standalone desktop GUI agent for executing computer tasks (e.g., web browsing, file management, application control)

How to Use

from transformers import Qwen3VLForConditionalGeneration, AutoProcessor

model = Qwen3VLForConditionalGeneration.from_pretrained(
    "UI-MOPD/Qwen3-VL-32B-Thinking-Desktop-Teacher",
    torch_dtype="auto",
    device_map="auto",
)
processor = AutoProcessor.from_pretrained("UI-MOPD/Qwen3-VL-32B-Thinking-Desktop-Teacher")

Citation

@misc{lian2026uimopdmultiplatformonpolicydistillation,
      title={UI-MOPD: Multi-Platform On-Policy Distillation for Continual GUI Agent Learning}, 
      author={Niu Lian and Alan Chen and Zhehao Yu and Chengzhen Duan and Fazhan Liu and Hui Liu and Pei Fu and Jian Luan and Yaowei Wang and Shu-Tao Xia and Jinpeng Wang},
      year={2026},
      eprint={2607.04425},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2607.04425}, 
}

Related Resources

• Project Page: https://elisspectre.github.io/UI-MOPD/
• Code: https://github.com/EliSpectre/UI-MOPD
• Paper: UI-MOPD (arXiv)