Salesforce/GTA1-7B

VISIONPricing:Input $0.4 / Output $0.8Concurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:32kPublished:Oct 1, 2025License:mitArchitecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Salesforce/GTA1-7B is a 7 billion parameter open-source GUI grounding model developed by Salesforce, designed for robust GUI automation and interaction. Utilizing Reinforcement Learning (RL) with GRPO, it excels at directly incentivizing actionable and grounded responses rather than relying on verbose Chain-of-Thought reasoning. This model achieves state-of-the-art performance on challenging GUI grounding benchmarks like ScreenSpot-V2 (93.4%), ScreenSpotPro (55.5%), OSWORLD-G (60.1%), and OSWORLD-G-Refined (68.8%), making it highly effective for tasks requiring precise GUI element identification and interaction.

Loading preview...

Overview

Salesforce/GTA1-7B is a 7 billion parameter open-source GUI grounding model developed by Salesforce, specifically designed for robust GUI automation and interaction. Unlike models that rely on extensive Chain-of-Thought (CoT) reasoning, GTA1-7B leverages Reinforcement Learning (RL) with GRPO to directly incentivize actionable and grounded responses, rewarding successful clicks. This approach leads to highly effective and precise GUI element identification and interaction.

Key Capabilities

  • State-of-the-Art GUI Grounding: Consistently achieves top results across challenging benchmarks, outperforming other open-source and even some proprietary models.
  • Reinforcement Learning (GRPO): Utilizes GRPO for objective alignment, directly rewarding successful GUI interactions.
  • High Accuracy on Benchmarks: Achieves 93.4% on ScreenSpot-V2, 55.5% on ScreenSpotPro, 60.1% on OSWORLD-G, and 68.8% on OSWORLD-G-Refined, demonstrating strong performance in identifying and interacting with GUI elements.
  • Agent Performance: Shows strong performance on OSWorld, OSWorld-Verified, and WindowsAgentArena benchmarks, indicating its capability in complex agentic tasks.

Good For

  • Automated GUI Testing: Ideal for creating agents that can accurately interact with graphical user interfaces for testing purposes.
  • Robotic Process Automation (RPA): Suitable for automating repetitive tasks that involve navigating and interacting with software applications.
  • Research in GUI Agents: Provides a strong baseline and tool for researchers exploring advanced GUI interaction and grounding techniques.
  • Developing Intelligent Assistants: Can be integrated into systems requiring precise visual understanding and interaction with digital interfaces.

Finetunes
2 models