Model reference · open weights

UI-TARS-SFT

Available as managed deployment LLMs ByteDance-Seed Vision + text 1 variants 5k dl/mo

UI-TARS-SFT is an open-weight language model from ByteDance-Seed. AxForge deploys and operates it for you on dedicated EU-owned hardware — with the licence handled where one is required.

Available as managed deployment — configured and operated for you on dedicated EU hardware, quoted per deployment.

What it is

MakerByteDance-Seed
TypeLanguage models
TaskVision + text
Parameters (lead)2.4B
Runs withtransformers
Released2025-01-20
Popularity5k downloads / month
LicenceOpen weights

About

What UI-TARS-SFT is

UI-TARS-2B-SFT  |  UI-TARS-2B-gguf  |  UI-TARS-7B-SFT  |  UI-TARS-7B-DPO(Recommended)  |  UI-TARS-7B-gguf  |  UI-TARS-72B-SFT  |  UI-TARS-72B-DPO(Recommended)

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

ModelVisualWebBenchWebSRCSQAshort
Qwen2-VL-7B73.381.884.9
Qwen-VL-Max74.191.178.6
Gemini-1.5-Pro75.488.982.2
UIX-Qwen2-7B75.982.978.8
Claude-3.5-Sonnet78.290.483.1
GPT-4o78.587.782.3
UI-TARS-2B72.989.286.4
UI-TARS-7B79.793.687.7
UI-TARS-72B82.889.388.6

Grounding Capability Evaluation

  • ScreenSpot Pro
Agent ModelDev-TextDev-IconDev-AvgCreative-TextCreative-IconCreative-AvgCAD-TextCAD-IconCAD-AvgScientific-TextScientific-IconScientific-AvgOffice-TextOffice-IconOffice-AvgOS-TextOS-IconOS-AvgAvg-TextAvg-IconAvg
QwenVL-7B0.00.00.00.00.00.00.00.00.00.70.00.40.00.00.00.00.00.00.10.00.1
GPT-4o1.30.00.71.00.00.62.00.01.52.10.01.21.10.00.90.00.00.01.30.00.8
SeeClick0.60.00.31.00.00.62.50.01.93.50.02.01.10.00.92.80.01.51.80.01.1
Qwen2-VL-7B2.60.01.31.50.00.90.50.00.46.30.03.53.41.93.00.90.00.52.50.21.6
OS-Atlas-4B7.10.03.73.01.42.32.00.01.59.05.57.55.13.84.85.60.03.15.01.73.7
ShowUI-2B16.91.49.49.10.05.32.50.01.913.27.310.615.37.513.510.32.26.610.82.67.7
CogAgent-18B14.90.78.09.60.05.67.13.16.122.21.813.413.00.010.05.60.03.112.00.87.7
Aria-UI16.20.08.423.72.114.77.61.66.127.16.418.120.31.916.14.70.02.617.12.011.3
UGround-7B26.62.114.727.32.817.014.21.611.131.92.719.331.611.327.017.80.09.725.02.816.5
Claude Computer Use22.03.912.625.93.416.814.53.711.933.915.825.830.116.326.911.04.58.123.47.117.1
OS-Atlas-7B33.11.417.728.82.817.912.24.710.337.57.324.433.95.727.427.14.516.828.14.018.9
UGround-V1-7B--35.5--27.8--13.5--38.8--48.8--26.1--31.1
UI-TARS-2B47.44.126.442.96.327.617.84.714.656.917.3

From the published model card. Full card on the HuggingFace links in the sidebar.

Using it via the API

Call it like any OpenAI endpoint

Once AxForge deploys ui-tars-sft for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (ui-tars-sft below is illustrative; you get the exact model name on deployment.)

$ curl -sS https://api.axforge.ai/v1/chat/completions \
  -H "Authorization: Bearer $AXFORGE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"ui-tars-sft","messages":[{"role":"user","content":"Hello"}]}'

Create an account — your API key is available in the console. 5M tokens/month currently included with every new account at launch.

© 2026 AxForge · EU-hosted AI infrastructure Pricing Docs Trust Privacy Terms