---
title: Indeed-UI-8B
canonical_url: "https://www.modelscope.cn/models/IntellgenceIndeed/Indeed-UI-8B"
md_url: "https://www.modelscope.cn/models/IntellgenceIndeed/Indeed-UI-8B.md"
repository: IntellgenceIndeed/Indeed-UI-8B
last_updated: 2026-08-20
pipeline_tag: visual-grounding
tasks:
  - visual-grounding
model_type:
  - qwen3_vl
architectures:
  - Qwen3VLForConditionalGeneration
parameters: 8.8B
tensor_type:
  - BF16
library_name:
  - safetensors
downloads: 27
stars: 1
---

# Indeed-UI-8B

> Indeed-UI-8B - IntellgenceIndeed 在 ModelScope 开源的模型。Indeed-UI-8B is an enhanced GUI agent model built upon GUI-Owl 1.5-8B. We further optimized the model with GUI grounding data and high-resolution training, with a particular focus on precise UI element…

- **Repository**: IntellgenceIndeed/Indeed-UI-8B
- **Tasks**: visual-grounding
- **Parameters**: 8.8B
- **Downloads**: 27
- **Stars**: 1
- **Last updated**: 2026-08-20

Source: https://www.modelscope.cn/models/IntellgenceIndeed/Indeed-UI-8B

---

# Indeed-UI-8B

**Indeed-UI-8B** is an enhanced GUI agent model built upon **GUI-Owl 1.5-8B**. We further optimized the model with GUI grounding data and high-resolution training, with a particular focus on precise UI element localization and robust visual grounding across different GUI environments.

## 📊 Benchmark Results

| **Benchmark**                | **Score**|
| ---------------------------- | -------- |
| **ScreenSpot-Pro**           | **73.4** |
| **ScreenSpot-Pro (Zoom-In)** | **81.0** |
| **ScreenSpot-V2**            | **94.4** |
| **OSWorld-G**                | **68.3** |
| **OSWorld-G-refined**        | **70.9** |

The results demonstrate strong GUI grounding performance across both standard and high-resolution GUI benchmarks.

## 🙏 Acknowledgements

Indeed-UI is built upon **GUI-Owl 1.5**. We sincerely thank the authors and contributors of GUI-Owl and the open-source GUI research community for their valuable contributions.

We also thank the creators of the benchmark datasets for making these resources publicly available and enabling reproducible research in GUI agents.