tencent/UI-Mate-27B

VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 14, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Cold

UI-Mate-27B is a 27 billion parameter open-weight foundation GUI agent developed by Tencent HY Frontier, built upon the Qwen3.6-27B base model. It is designed for long-horizon computer use, observing live screenshots and generating structured keyboard and mouse actions for native desktop interaction. This model excels at general computer-use performance across Ubuntu and Windows benchmarks, planning and re-planning tasks based on real-time screen changes.

Loading preview...

Overview of UI-Mate-27B

UI-Mate-27B, developed by Tencent HY Frontier, is a 27 billion parameter open-weight foundation GUI agent built on the Qwen3.6-27B base model. It is engineered for long-horizon tasks across various applications and operating systems. The model processes natural-language instructions, live screenshots, and interaction history to produce reasoning, action descriptions, and structured computer-use tool calls, effectively interacting with native desktop environments.

Key Capabilities

  • Live-Screen Grounding: Operates by observing live screenshots and adapting to interface changes, rather than relying on pre-recorded coordinates.
  • Structured Actions: Generates mouse, keyboard, scrolling, waiting, and user interaction actions compatible with pyautogui.
  • Long-Horizon Execution: Capable of performing complex tasks spanning multiple applications.
  • Robust Performance: Demonstrates strong general computer-use performance across Ubuntu and Windows benchmarks, achieving scores like 77.0 on OSWorld-Verified and 66.2 on WindowsAgentArena.
  • Reinforcement Learning: Utilizes supervised fine-tuning followed by online reinforcement learning in executable GUI environments for enhanced adaptability.

Good For

  • Research and Development: Ideal for advancing screenshot-based GUI agents in controlled desktop environments.
  • Automating Desktop Tasks: Suitable for tasks requiring interaction with native desktop applications based on visual input.
  • Building GUI Agents: Provides a robust foundation for creating agents that can plan and execute actions from live screen observations.

It's important to note that UI-Mate-27B is an agent checkpoint and requires an external runtime and the official prompt, response parser, and interaction harness from the UI-Mate repository for optimal use. Users should prioritize safety by using isolated environments and human confirmation for sensitive operations.