inclusionAI/UI-Venus-Navi-72B

VISIONPricing:Input $3.2 / Output $4Concurrent Unit Cost:4Model Size:72BQuant:FP8Context Size:32kPublished:Aug 16, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

UI-Venus-Navi-72B by inclusionAI is a 72-billion parameter multimodal UI agent built on Qwen2.5-VL, designed for precise GUI element grounding and effective navigation using only screenshots. It achieves state-of-the-art performance on benchmarks like Screenspot-v2/Pro and AndroidWorld through Reinforcement Fine-Tuning (RFT) and high-quality data. This model excels at visual grounding, UI navigation, cross-platform generalization, and complex task reasoning, making it suitable for autonomous UI interaction and agent-based systems.

Loading preview...

UI-Venus-Navi-72B: High-Performance UI Agent

UI-Venus-Navi-72B, developed by inclusionAI, is a 72-billion parameter multimodal UI agent built upon the Qwen2.5-VL architecture. It specializes in performing accurate GUI element grounding and effective navigation using only screenshots as input. 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

  • Screenshot-based UI Interaction: Operates solely on visual input (screenshots) for GUI element grounding and navigation.
  • State-of-the-Art Performance: Achieves 95.3% on Screenspot-Pro and 65.9% success rate on AndroidWorld, surpassing previous models like GTA1 and UI-TARS-1.5.
  • Reinforcement Fine-Tuning (RFT): Utilizes reward functions for both grounding and navigation tasks, enhancing performance.
  • Efficient Data Cleaning: Trained on hundreds of thousands of high-quality samples for robustness.
  • Advanced Reasoning: Incorporates Self-Evolving Trajectory History Alignment and Sparse Action Enhancement for improved reasoning coherence and long-horizon planning.

Good for

  • Autonomous UI Interaction: Developing agents that can interact with graphical user interfaces without explicit programming.
  • Cross-Platform UI Automation: Tasks requiring navigation and interaction across different UI environments (e.g., web, mobile).
  • Research in UI Agents: Advancing the field of agent-based systems and multimodal understanding for UI.
  • Complex Task Reasoning: Scenarios demanding strong planning and generalization capabilities for multi-step UI tasks.

Finetunes
1 models