Tencent has released the UI-Mate-democua-27B checkpoint, a demonstration-guided version of its UI-Mate GUI agent that can adapt recorded workflows to new tasks. The model observes live screenshots and reasons over visible states to produce structured actions, using demonstrations as guidance rather than fixed scripts.
- Parameters: 27B, based on Qwen3.6-27B, licensed under Apache-2.0.
- Trained via supervised fine-tuning on mixed general and demonstration-augmented data, preserving instruction-only competence.
- Demonstrations are segmented into subtasks with completion criteria; the model replans from live screens rather than replaying coordinates.
- Evaluation shows improved performance on OSWorld-Subset (18 tasks improved) and GameDev (shorter trajectories), with perfect scores on previously unsolvable tasks like chrome-02.
The release enables research into screenshot-based GUI agents by allowing models to learn procedural knowledge from in-context demonstrations, improving success rates on complex multi-application tasks.