Salesforce/GTA1-7B-2507
Salesforce/GTA1-7B-2507 is a 7 billion parameter GUI grounding model developed by Salesforce, trained using Reinforcement Learning (GRPO) to directly incentivize actionable and grounded responses. It excels at identifying GUI elements and performing automated tasks, achieving state-of-the-art performance on challenging datasets like ScreenSpot-V2, ScreenSpotPro, OSWORLD-G, and WindowsAgentArena. This model is optimized for robust GUI automation and agent performance, providing precise coordinate predictions for user interface elements.
Loading preview...
Overview
Salesforce/GTA1-7B-2507 is a 7 billion parameter GUI grounding model developed by Salesforce, designed to enhance automated interaction with graphical user interfaces. Unlike models that rely on verbose Chain-of-Thought (CoT) reasoning, GTA1-7B-2507 utilizes Reinforcement Learning (GRPO) to directly align with objectives by rewarding successful actions, leading to more grounded and actionable responses. This approach makes it highly effective for tasks requiring precise GUI element identification and interaction.
Key Capabilities
- State-of-the-Art GUI Grounding: Achieves top performance across challenging datasets including ScreenSpot-V2 (92.4%), ScreenSpotPro (50.1%), and OSWORLD-G (55.1%), demonstrating significant improvements over baseline models.
- Robust Agent Performance: Shows strong results on agent benchmarks like OSWorld (45.2%) and OSWorld-Verified (53.1%) when paired with models like OpenAI CUA o3, and excels in Windows-specific automation on the WindowsAgentArena benchmark (47.9% success rate).
- Precise Element Location: Designed to output exact (x,y) coordinates for specified GUI elements, facilitating accurate automated interactions.
- Reinforcement Learning Optimization: Leverages GRPO for objective alignment, directly incentivizing successful clicks and grounded responses rather than lengthy textual reasoning.
When to Use This Model
- Automated GUI Testing: Ideal for scenarios requiring automated interaction with user interfaces for testing purposes.
- Robotic Process Automation (RPA): Suitable for developing agents that need to navigate and interact with applications through their graphical interfaces.
- Research in GUI Grounding: A strong baseline and comparison point for further research into improving GUI understanding and interaction.
Important Considerations
Users should be aware of potential accuracy limitations and the need for human oversight, especially in production environments, to ensure compliance and prevent unintended actions. Thorough testing in controlled environments is recommended.