inclusionAI/UI-Venus-Navi-7B
UI-Venus-Navi-7B by inclusionAI is a 7 billion parameter native UI agent built on the Qwen2.5-VL multimodal large language model, designed for precise GUI element grounding and effective navigation using only screenshots. It achieves state-of-the-art performance on UI understanding and interaction benchmarks like Screenspot-v2/Pro and AndroidWorld through Reinforcement Fine-Tuning (RFT). This model excels at autonomous UI interaction and agent-based systems, offering strong planning and generalization capabilities for complex tasks.
Loading preview...
Overview
inclusionAI's UI-Venus-Navi-7B is a 7 billion parameter native UI agent, leveraging the Qwen2.5-VL multimodal large language model. It is specifically engineered for accurate GUI element grounding and effective navigation, operating solely on screenshot inputs. The model achieves state-of-the-art performance on key UI understanding and interaction benchmarks, including ScreenSpot-Pro, ScreenSpot-v2, OS-World-G, UI-Vision, and Android World.
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 reward functions for both grounding and navigation tasks, trained on hundreds of thousands of high-quality samples.
- High Performance: Achieves 94.1% on Screenspot-V2 and 49.1% success rate on AndroidWorld, outperforming many prior models.
- Advanced Reasoning: Incorporates Self-Evolving Trajectory History Alignment & Sparse Action Enhancement for improved reasoning coherence and long-horizon planning.
Good for
- Developing autonomous UI agents that interact with graphical user interfaces.
- Tasks requiring precise visual grounding and navigation based on screenshots.
- Research and development in agent-based systems and UI automation.