inclusionAI/UI-Venus-1.5-8B
inclusionAI/UI-Venus-1.5-8B is an 8 billion parameter unified, end-to-end GUI Agent developed by inclusionAI, designed for robust real-world applications. Built upon the Qwen3-VL Series, this model excels in grounding and agent benchmarks, achieving state-of-the-art performance on tasks like ScreenSpot-Pro, VenusBench-GD, and AndroidWorld. It is specifically optimized for complex GUI navigation and interaction across mobile and web environments, demonstrating strong generalization capabilities.
Loading preview...
UI-Venus-1.5-8B: A Unified GUI Agent
UI-Venus-1.5-8B is an 8 billion parameter model from the UI-Venus 1.5 family, developed by inclusionAI as a unified, end-to-end GUI Agent. This model is engineered for robust real-world applications, focusing on interaction and navigation within graphical user interfaces.
Key Capabilities & Features
- Unified GUI Agent: Designed to handle diverse GUI tasks across mobile and web platforms.
- Advanced Training Pipeline: Utilizes a four-stage progressive training curriculum, starting from the Qwen3-VL Series, incorporating Mid-Training with large-scale GUI data, Offline-RL for task-specific optimization, Online-RL for enhanced navigation, and Model Merging to unify specialized components.
- State-of-the-Art Performance: Achieves leading results on key grounding benchmarks such as ScreenSpot-Pro (68.4% for 8B), VenusBench-GD, OSWorld-G, and UI-Vision. It also demonstrates superior performance on agent benchmarks like AndroidWorld (77.6% for 30B-A3B, with 8B showing significant gains over previous generations), AndroidLab, and WebVoyager.
- Robust Navigation: Excels in complex, real-world scenarios, including robust navigation across 40+ Chinese mobile apps.
- Efficient Scaling: The 8B variant significantly outperforms previous 72B models on Android Lab and VenusBench-Mobile, showcasing the effectiveness of its updated training methodology.
- Cross-Platform Generalization: Exhibits strong adaptability across different operating systems and input modalities, performing well in programmatic Android environments and dynamic web navigation.
Ideal Use Cases
- Automated GUI Interaction: Developing agents for automating tasks within mobile applications and web browsers.
- UI Testing and Validation: Creating intelligent systems to test and validate user interfaces.
- Robotic Process Automation (RPA): Implementing agents for complex, multi-step processes involving GUI elements.
- Accessibility Tools: Building advanced tools that can interpret and interact with graphical interfaces for users with specific needs.
- Research in GUI Agents: A strong baseline or component for further research and development in unified GUI agents.