What are the hardware requirements for UI TARS 72B SFT?

UI TARS 72B SFT requires approximately 204.2 GB of VRAM and supports a context window of 32K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed UI TARS 72B SFT and how large is it?

UI TARS 72B SFT is developed by ByteDance-Seed, a model with 72b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

Where can I download or evaluate UI TARS 72B SFT?

UI TARS 72B SFT is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

UI TARS 72B SFT by ByteDance-Seed — VRAM 204.2GB, 32K context

Name: UI TARS 72B SFT
Author: ByteDance-Seed

UI TARS 72B SFT is an open-source language model by ByteDance-Seed. Features: 72b LLM, VRAM: 204.2GB, Context: 32K, License: apache-2.0, LLM Explorer Score: 0.17.

Arxiv:2501.12326 Conversational En Endpoints compatible Eval-results Gui Image-text-to-text Multimodal Qwen2 vl Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/ByteDance-Seed/UI-TARS-72B-SFT

UI TARS 72B SFT Parameters and Internals

LLM Name	UI TARS 72B SFT
Repository 🤗	https://huggingface.co/ByteDance-Seed/UI-TARS-72B-SFT
Model Size	72b
Required VRAM	204.2 GB
Updated	2026-08-08
Maintainer	ByteDance-Seed
Model Type	qwen2_vl
Model Files	2.8 GB: 1-of-64 5.0 GB: 2-of-64 4.1 GB: 3-of-64 4.5 GB: 4-of-64 4.5 GB: 5-of-64 4.8 GB: 6-of-64 4.8 GB: 7-of-64 4.5 GB: 8-of-64 4.8 GB: 9-of-64 4.8 GB: 10-of-64 4.5 GB: 11-of-64 4.8 GB: 12-of-64 4.8 GB: 13-of-64 4.5 GB: 14-of-64 4.8 GB: 15-of-64 4.8 GB: 16-of-64 4.5 GB: 17-of-64 4.8 GB: 18-of-64 4.8 GB: 19-of-64 4.5 GB: 20-of-64 4.8 GB: 21-of-64 4.8 GB: 22-of-64 4.5 GB: 23-of-64 4.8 GB: 24-of-64 4.8 GB: 25-of-64 4.5 GB: 26-of-64 4.8 GB: 27-of-64 4.8 GB: 28-of-64 4.5 GB: 29-of-64 4.8 GB: 30-of-64 4.8 GB: 31-of-64 4.5 GB: 32-of-64 4.8 GB: 33-of-64 4.8 GB: 34-of-64 4.5 GB: 35-of-64 4.8 GB: 36-of-64 4.8 GB: 37-of-64 4.5 GB: 38-of-64 4.8 GB: 39-of-64 4.8 GB: 40-of-64 4.5 GB: 41-of-64 4.8 GB: 42-of-64 4.8 GB: 43-of-64 4.5 GB: 44-of-64
Supported Languages	en
Model Architecture	Qwen2VLForConditionalGeneration
License	apache-2.0
Context Length	32768
Model Max Length	32768
Transformers Version	4.41.2
Vocabulary Size	152064
Torch Data Type	bfloat16

Rank the UI TARS 72B SFT capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to UI TARS 72B SFT

Best Alternatives	Context / RAM	Downloads	Likes
QVQ 72B Preview Bf16	125K / 147.7 GB	23	3
UI TARS 72B DPO	32K / 147.5 GB	1735	158
UI TARS 72B DPO Bf16	32K / 147.7 GB	30	1
UI TARS 72B SFT Bf16	32K / 147.7 GB	32	1
QVQ 72B Preview 4bit	125K / 41.3 GB	20	7
QVQ 72B Preview 6bit	125K / 59.7 GB	34	2
QVQ 72B Preview 3bit	125K / 32.2 GB	21	5
QVQ 72B Preview 8bit	125K / 77.8 GB	22	3
Qwen2 VL 72B Instruct 8bit	32K / 77.8 GB	79	4
Qwen2 VL 72B Instruct 4bit	32K / 41.3 GB	72	5

Note: green Score (e.g. "73.2") means that the model is better than ByteDance-Seed/UI-TARS-72B-SFT.