Tencent HY Frontier Multimodal Agent Team has released UI-Mate-27B, an open-weight foundation GUI agent designed for long-horizon tasks across applications and operating systems. The model observes live screenshots to reason over visible states and produces structured keyboard and mouse actions for native desktop interaction.
- Base model: Qwen3.6-27B with 27 billion parameters.
- Supports general computer use from natural-language instructions and demonstration-guided use that adapts reusable workflows.
- Training involves supervised fine-tuning followed by online reinforcement learning in executable GUI environments.
- Demonstrations are treated as guidance, allowing the model to re-plan when interface states differ.
- On GameDev benchmarks, demonstration guidance reduces average trajectory length from 303.6 to 253.1 steps.
The release provides an Apache-2.0 licensed checkpoint compatible with vLLM and OpenAI interfaces, intended for research and development in controlled desktop environments.