UI-MOPD/Qwen3-VL-32B-Thinking-Mobile-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-Mobile-Teacher is a 33.4 billion parameter multimodal model developed by UI-MOPD, fine-tuned from Qwen3-VL-32B-Thinking. It specializes in mobile GUI interaction tasks, serving as a platform-specific expert within the UI-MOPD multi-teacher distillation framework. This model excels at executing mobile tasks like app navigation and settings control, achieving a 16.2% task success rate on MobileWorld benchmarks.

Loading preview...

Qwen3-VL-32B-Thinking-Mobile-Teacher Overview

This model is the Mobile Teacher component of the UI-MOPD project, a platform-specific expert designed for mobile Graphical User Interface (GUI) interaction tasks. It is fine-tuned from the Qwen3-VL-32B-Thinking base model, leveraging approximately 160,000 mobile interaction steps from the Uni-GUI dataset.

Key Capabilities and Role

  • Specialized Mobile GUI Agent: Functions as a standalone agent for tasks such as app navigation, settings control, and messaging on mobile platforms.
  • Teacher Model in Distillation: Primarily designed to serve as a teacher in the UI-MOPD multi-teacher on-policy distillation framework, guiding a smaller 8B student policy.
  • Enhanced Mobile Performance: Achieves a 16.2% task success rate on the MobileWorld benchmark, significantly outperforming its base model's 9.4% success rate.

Training and Intended Use

This model represents Stage 1 of the UI-MOPD training pipeline, focusing on supervised fine-tuning for mobile environments. Its primary purpose is to enable the development of cross-platform GUI agents by providing expert guidance for mobile-specific interactions. Developers can also utilize it directly for automating mobile tasks.

Finetunes
1 models