inclusionAI/UI-Venus-Ground-72B
inclusionAI/UI-Venus-Ground-72B is a 72 billion parameter multimodal UI agent built on the Qwen2.5-VL architecture, developed by inclusionAI. This model is specifically designed for high-performance GUI element grounding and effective navigation using only screenshots as input. It achieves state-of-the-art results on UI understanding and interaction benchmarks like ScreenSpot-v2 and ScreenSpot-Pro, making it ideal for autonomous UI interaction and agent-based systems.
Loading preview...
UI-Venus-Ground-72B: High-Performance UI Agent
UI-Venus-Ground-72B is a 72 billion parameter multimodal UI agent developed by inclusionAI, based on the Qwen2.5-VL architecture. It specializes in precise GUI element grounding and effective navigation, operating solely from screenshots. The model achieves state-of-the-art performance across various UI understanding and interaction benchmarks, including ScreenSpot-v2, ScreenSpot-Pro, OS-World-G, UI-Vision, and AndroidWorld.
Key Capabilities
- State-of-the-Art UI Agent: Publicly released to advance research in autonomous UI interaction and agent-based systems.
- Reinforcement Fine-Tuning (RFT): Utilizes carefully designed reward functions for both grounding and navigation tasks, contributing to its superior performance.
- Efficient Data Cleaning: Trained on hundreds of thousands of high-quality samples for robustness.
- Advanced Planning: Incorporates Self-Evolving Trajectory History Alignment and Sparse Action Enhancement for improved reasoning coherence and long-horizon planning.
- Benchmark Performance: Achieves 95.3% on ScreenSpot-v2 and 61.9% on ScreenSpot-Pro, surpassing previous state-of-the-art models like GTA1 and UI-TARS-1.5.
Good For
- Developing autonomous UI agents.
- Applications requiring precise GUI element identification and interaction.
- Research in multimodal AI for human-computer interaction.
- Tasks involving UI navigation and complex task reasoning from visual input.