inclusionAI/UI-Venus-2-9B

Hugging Face
VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 26, 2026Architecture:Transformer0.0K Featherless Exclusive Warm

UI-Venus-2-9B by inclusionAI is a 9 billion parameter general-purpose foundation GUI agent, initialized from Qwen3.5-9B, designed for unified closed-loop reasoning and action across mobile applications, web platforms, and desktop operating systems. It features an expanded executable mobile pool of over 170 multilingual apps, 4,000+ web domains, and native desktop OS support. The model excels in GUI grounding, mobile, web, and computer-use benchmarks, and demonstrates improved safety with an 11.3% Attack Success Rate on OSHarm.

Loading preview...

UI-Venus-2-9B: A Unified GUI Agent

UI-Venus-2-9B is a 9 billion parameter general-purpose foundation GUI agent developed by inclusionAI, built upon the Qwen3.5-9B base model. It is engineered for comprehensive interaction across diverse digital environments, including mobile applications, web platforms, and desktop operating systems, utilizing a closed-loop reasoning-action framework.

Key Capabilities & Features

  • Broad Environmental Coverage: Operates across 170+ multilingual mobile apps (Chinese and English), 4,000+ web domains, and native desktop OS environments.
  • Robust Task Generation: Employs a deep-research pipeline to generate accurate and executable task queries grounded in application functionality.
  • Advanced Verification: Utilizes trace-level and sample-level evaluators based on visual keypoints and multi-model voting for reliable reward signals, robust against reward hacking.
  • Enhanced Safety: Achieves a significantly lower Attack Success Rate (ASR) on safety benchmarks like OSHarm (11.3%) and OSBlind (48.8%) compared to its base model.
  • Strong Performance: Demonstrates near state-of-the-art results in GUI grounding, mobile, web, computer-use, and CAPTCHA benchmarks among models of comparable scale.

Good For

  • Developing automated GUI agents for mobile, web, and desktop applications.
  • Tasks requiring precise visual grounding and interaction with user interfaces.
  • Applications demanding robust performance in CAPTCHA solving and complex multi-step operations across different platforms.
  • Use cases where safety and reduced susceptibility to harmful actions in GUI environments are critical.

Finetunes
2 models