bytedance-research

UI-TARS-7B-DPO

Deploy Dedicated

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Get help setting up a custom Dedicated Endpoints.

Talk with our engineer to get a quote for reserved GPU instances with discounts.

Introduction

UI-TARS is a next-generation native GUI agent model designed to interact seamlessly with graphical user interfaces (GUIs) using human-like perception, reasoning, and action capabilities. Unlike traditional modular frameworks, UI-TARS integrates all key components—perception, reasoning, grounding, and memory—within a single vision-language model (VLM), enabling end-to-end task automation without predefined workflows or manual rules.

This repository contains the model for the paper UI-TARS: Pioneering Automated GUI Interaction with Native Agents.

Code: https://github.com/bytedance/UI-TARS

Performance

Perception Capabilty Evaluation

Table with columns: Model, VisualWebBench, WebSRC, SQAshort
Model	VisualWebBench	WebSRC	SQAshort
Qwen2-VL-7B	73.3	81.8	84.9
Qwen-VL-Max	74.1	91.1	78.6
Gemini-1.5-Pro	75.4	88.9	82.2
UIX-Qwen2-7B	75.9	82.9	78.8
Claude-3.5-Sonnet	78.2	90.4	83.1
GPT-4o	78.5	87.7	82.3
UI-TARS-2B	72.9	89.2	86.4
UI-TARS-7B	79.7	93.6	87.7
UI-TARS-72B	82.8	89.3	88.6

Grounding Capability Evaluation

ScreenSpot Pro

Table with columns: Agent Model, Dev-Text, Dev-Icon, Dev-Avg, Creative-Text, Creative-Icon, Creative-Avg, CAD-Text, CAD-Icon, CAD-Avg, Scientific-Text, Scientific-Icon, Scientific-Avg, Office-Text, Office-Icon, Office-Avg, OS-Text, OS-Icon, OS-Avg, Avg-Text, Avg-Icon, Avg
Agent Model	Dev-Text	Dev-Icon	Dev-Avg	Creative-Text	Creative-Icon	Creative-Avg	CAD-Text	CAD-Icon	CAD-Avg	Scientific-Text	Scientific-Icon	Scientific-Avg	Office-Text	Office-Icon	Office-Avg	OS-Text	OS-Icon

ScreenSpot

Table with columns: Method, Mobile-Text, Mobile-Icon/Widget, Desktop-Text, Desktop-Icon/Widget, Web-Text, Web-Icon/Widget, Avg
Method	Mobile-Text	Mobile-Icon/Widget	Desktop-Text	Desktop-Icon/Widget	Web-Text	Web-Icon/Widget	Avg
Agent Framework
GPT-4 (SeeClick)	76.6	55.5	68.0

ScreenSpot v2

Table with columns: Method, Mobile-Text, Mobile-Icon/Widget, Desktop-Text, Desktop-Icon/Widget, Web-Text, Web-Icon/Widget, Avg
Method	Mobile-Text	Mobile-Icon/Widget	Desktop-Text	Desktop-Icon/Widget	Web-Text	Web-Icon/Widget	Avg
Agent Framework
GPT-4o (SeeClick)	85.2	58.8	79.9

Offline Agent Capability Evaluation

Multimodal Mind2Web

Table with columns: Method, Cross-Task Ele.Acc, Cross-Task Op.F1, Cross-Task Step SR, Cross-Website Ele.Acc, Cross-Website Op.F1, Cross-Website Step SR, Cross-Domain Ele.Acc, Cross-Domain Op.F1, Cross-Domain Step SR
Method	Cross-Task Ele.Acc	Cross-Task Op.F1	Cross-Task Step SR	Cross-Website Ele.Acc	Cross-Website Op.F1	Cross-Website Step SR	Cross-Domain Ele.Acc	Cross-Domain Op.F1	Cross-Domain Step SR
Agent Framework

Android Control and GUI Odyssey

Table with columns: Agent Models, AndroidControl-Low Type, AndroidControl-Low Grounding, AndroidControl-Low SR, AndroidControl-High Type, AndroidControl-High Grounding, AndroidControl-High SR, GUIOdyssey Type, GUIOdyssey Grounding, GUIOdyssey SR
Agent Models	AndroidControl-Low Type	AndroidControl-Low Grounding	AndroidControl-Low SR	AndroidControl-High Type	AndroidControl-High Grounding	AndroidControl-High SR	GUIOdyssey Type	GUIOdyssey Grounding	GUIOdyssey SR
Claude	74.3	0.0	19.4	63.7	0.0	12.5	60.9	0.0

Online Agent Capability Evaluation

Table with columns: Method, OSWorld (Online), AndroidWorld (Online)
Method	OSWorld (Online)	AndroidWorld (Online)
Agent Framework
GPT-4o (UGround)	-	32.8
GPT-4o (Aria-UI)	15.2	44.8
GPT-4o (Aguvis-7B)	14.8	37.1
GPT-4o (Aguvis-72B)	17.0	-
GPT-4o (OS-Atlas-7B)	14.6

Citation

If you find our paper and model useful in your research, feel free to give us a cite.

BibTeX
@article{qin2025ui,
  title={UI-TARS: Pioneering Automated GUI Interaction with Native Agents},
  author={Qin, Yujia and Ye, Yining and Fang, Junjie and Wang, Haoming and Liang, Shihao and Tian, Shizuo and Zhang, Junda and Li, Jiahao and Li, Yunxin and Huang, Shijue and others},
  journal={arXiv preprint arXiv:2501.12326},
  year={2025}
}

Model provider

bytedance-research

Model tree

Base

this model

Modalities

Input

Text, Image

Output

Text

Pricing

Dedicated Endpoints

View details

Supported Functionality

Model APIs

Dedicated Endpoints

Container

More information

Model card