tencent/UI-Mate-9B

VISIONConcurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 14, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

UI-Mate-9B by Tencent HY Frontier is a 9 billion parameter open-weight foundation GUI agent based on Qwen3.5-9B, designed for long-horizon desktop interaction. It processes live screenshots and natural language instructions to generate structured keyboard and mouse actions. This model excels at general computer use by reasoning over visible screen states and adapting workflows across applications and operating systems.

Loading preview...

Overview of UI-Mate-9B

UI-Mate-9B, developed by Tencent HY Frontier, is a 9 billion parameter open-weight foundation GUI agent built upon the Qwen3.5-9B base model. It is specifically engineered for long-horizon tasks across various applications and operating systems, functioning by observing live screenshots and interpreting natural language instructions. The model then generates structured keyboard and mouse actions to interact with the graphical user interface.

Key Capabilities

  • Live-Screen Grounding: Reasons over current screen states rather than relying on pre-recorded coordinate replays.
  • Structured Actions: Produces actions compatible with pyautogui for precise control.
  • Long-Horizon Execution: Capable of performing complex tasks spanning multiple applications.
  • Efficient Scale: Offers strong general computer-use performance at an efficient 9B parameter count.
  • Adaptive Planning: Plans actions based on the visible interface and re-plans as the UI changes.

Evaluation and Use Cases

UI-Mate-9B demonstrates robust performance on GUI agent benchmarks, achieving scores of 66.2 on OSWorld-Verified, 61.7 on WindowsAgentArena, and 34.00 (strict success) on OSWorkerBench. It is intended for research and development of screenshot-based GUI agents in controlled desktop environments. Users should note that its behavior can be influenced by application versions, screen layouts, and display scaling, and it requires an external runtime to execute its predicted actions. For optimal use, Tencent recommends using the official prompt, response parser, and interaction harness from the UI-Mate repository.

Finetunes
3 models