---
title: ScaleCUA-qwen3-vl-scienceboard-sft
canonical_url: "https://www.modelscope.cn/models/extreme1228/ScaleCUA-qwen3-vl-scienceboard-sft"
md_url: "https://www.modelscope.cn/models/extreme1228/ScaleCUA-qwen3-vl-scienceboard-sft.md"
repository: extreme1228/ScaleCUA-qwen3-vl-scienceboard-sft
last_updated: 2026-07-09
license: other
pipeline_tag: image-text-to-text
tasks:
  - image-text-to-text
model_type:
  - qwen3_vl
architectures:
  - Qwen3VLForConditionalGeneration
parameters: 8.8B
tensor_type:
  - BF16
library_name:
  - safetensors
downloads: 37
stars: 0
tags:
  - computer-use-agent
  - gui-agent
  - scienceboard
  - scalecua
---

# ScaleCUA-qwen3-vl-scienceboard-sft

> ScaleCUA-qwen3-vl-scienceboard-sft - extreme1228 在 ModelScope 开源的模型。ScaleCUA-qwen3-vl-scienceboard-sft

extreme1228/ScaleCUA-qwen3-vl-scienceboard-sft 是 ModelScope 魔搭社区上的 8.8B 参数image-text-to-text模型，采用 other 许可。

- **Repository**: extreme1228/ScaleCUA-qwen3-vl-scienceboard-sft
- **License**: other
- **Tasks**: image-text-to-text
- **Parameters**: 8.8B
- **Tags**: computer-use-agent, gui-agent, scienceboard, scalecua
- **Downloads**: 37
- **Stars**: 0
- **Last updated**: 2026-07-09

Source: https://www.modelscope.cn/models/extreme1228/ScaleCUA-qwen3-vl-scienceboard-sft

---

# ScaleCUA-qwen3-vl-scienceboard-sft

A **Qwen3-VL-8B** computer-use agent from **ScaleCUA**, trained for **ScienceBoard**.
Stage: **SFT start-point**.

- 📄 Paper / code: [https://github.com/THUDM/SCALE-CUA](https://github.com/THUDM/SCALE-CUA)
- 🤗 Collection: [https://huggingface.co/collections/extreme1228/scalecua-6a4cad5e3c80a230622eaff3](https://huggingface.co/collections/extreme1228/scalecua-6a4cad5e3c80a230622eaff3) · 🤖 [ModelScope](https://modelscope.cn/collections/extreme1228/ScaleCUA)

## What is this

This is the **SFT start-point checkpoint** used to initialize online RL; the corresponding RL checkpoint is the released agent (see the collection).

ScaleCUA scales online RL for computer-use agents by pairing **VeriGen**
(verifiable GUI-task synthesis) with **Frontier Sampling** and **Visual Context
Segmentation**. See the repo for details.

## Usage

Serve with an OpenAI-compatible server (e.g. vLLM) and drive it with the
evaluation runtime in [`osworld_eval`](https://github.com/THUDM/SCALE-CUA/tree/main/osworld_eval):

```bash
pip install -U vllm
python -m vllm.entrypoints.openai.api_server --model extreme1228/ScaleCUA-qwen3-vl-scienceboard-sft \
  --served-model-name scalecua --trust-remote-code --port 8000
```

Then follow [`REPRODUCE.md`](https://github.com/THUDM/SCALE-CUA/blob/main/REPRODUCE.md) to run ScienceBoard evaluation.

## Results

 (see the paper for the full table).

## License

Weights are released for research; the model inherits the license of its base
model (**Qwen3-VL-8B**). Repository code is under the licenses in the GitHub repo.

## Citation

```bibtex
@misc{lv2026scalecua,
  title  = {ScaleCUA: Scaling Computer Use Agents with Verifiable Task Synthesis and Efficient Online RL},
  author = {Bowen Lv and Xiao Liu and Yanyu Ren and Hanyu Lai and Bohao Jing and Hanchen Zhang and Yanxiao Zhao and Shuntian Yao and Jie Tang and Yuxiao Dong},
  year   = {2026},
  url    = {https://github.com/THUDM/SCALE-CUA}
}
```