UI-MOPD/Qwen3-VL-32B-Thinking-Desktop-Teacher

VISIONConcurrent Unit Cost:2Model Size:33.4BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 4, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

UI-MOPD/Qwen3-VL-32B-Thinking-Desktop-Teacher is a 33.4 billion parameter multimodal model fine-tuned from Qwen3-VL-32B-Thinking. Developed by UI-MOPD, it specializes in desktop GUI interaction tasks, serving as a platform-specific expert. This model is designed as a teacher in the UI-MOPD distillation framework and can also function as a standalone desktop GUI agent for tasks like web browsing and file management.

Loading preview...

Model Overview

UI-MOPD/Qwen3-VL-32B-Thinking-Desktop-Teacher is a 33.4 billion parameter multimodal model, fine-tuned from Qwen3-VL-32B-Thinking. It is specifically designed as a Desktop Teacher model within the UI-MOPD project, focusing on desktop Graphical User Interface (GUI) interaction tasks.

Key Capabilities

  • Desktop GUI Expertise: Specialized in understanding and executing tasks within desktop environments, trained on the desktop subset of the Uni-GUI dataset, comprising approximately 160K interaction steps across 11.5K trajectories.
  • Teacher Model Role: Functions as a platform-specific teacher in the UI-MOPD multi-teacher on-policy distillation framework, guiding a smaller 8B student policy for cross-platform GUI agents.
  • Enhanced Performance: Achieves a 46.3% task success rate on OSWorld benchmarks, outperforming its base model (Qwen3-VL-32B-Thinking) which scored 41.0%.
  • Standalone GUI Agent: Can be used independently to automate computer tasks such as web browsing, file management, and application control.

Intended Use Cases

  • Training Cross-Platform GUI Agents: Ideal for researchers and developers working on distillation frameworks for GUI agents, leveraging its expertise as a desktop teacher.
  • Automating Desktop Tasks: Suitable for applications requiring automated interaction with desktop interfaces, offering a specialized solution for complex GUI operations.

Finetunes
1 models