Model reference · open weights

UI-TARS-1.5

LLMs ByteDance-Seed Vision + text 1 build Open weights 741k dl/mo

UI-TARS-1.5 is an open-weight language model from ByteDance. UI-TARS-1.5-7B (FP32) weighs 16.6 GB; the smallest configuration that runs it is 2× RTX 3060 12 GB.

UI-TARS-1.5 is an open-source multimodal agent developed by ByteDance for performing diverse tasks within virtual worlds, including computer use, browser operations, and phone interactions.
The model utilizes a vision-language architecture with reinforcement learning to enhance reasoning capabilities and is available under the Apache 2.0 license.
The released 7B variant contains 8.3B parameters, supports English, and focuses on general computer use capabilities.

Summary of the ByteDance-Seed/UI-TARS-1.5-7B model card, 2026-10-01

Hugging Face ↗GitHub ↗Paper ↗

What it is

Released by	ByteDance
Released	2025-04-16
Parameters	8.3B
VRAM	16.6 GB for the weights

What it runs on

Memory and cards for UI-TARS-1.5-7B (FP32)

16.6 GBweights, file size

57 MBcache per 1K tokens

2.1 GBruntime overhead, at least

128,000 tokenscontext max

Card	Requests at once		Context max	Memory
Card	8K each	32K each	Context max	Memory
RTX 3060 12 GB … RTX 4060 Ti 16 GB 2 smaller cards	—	—	—
RTX 3090 24 GB	10	2	80K	23.4 GB
RTX 4090 24 GB	9	2	79K	23.4 GB
RTX 5090 32 GB	26	6	all 125K	31.0 GB
L40S 48 GB	53	13	all 125K	44.0 GB
A100 80 GB	126	31	all 125K	78.2 GB
H100 80 GB	117	29	all 125K	78.1 GB
RTX PRO 6000 Blackwell 96 GB	150	37	all 125K	93.8 GB
DGX Spark (GB10) 128 GB unified	179	44	all 125K	107 GB
H200 141 GB	244	61	all 125K	138 GB
B200 180 GB	323	80	all 125K	176 GB
2× RTX 3060 12 GB tensor parallel	5	1	42K	11.6 GB a card
2× RTX 4060 Ti 16 GB tensor parallel	21	5	all 125K	15.4 GB a card

Memory needed at each load

Requests at once	8K tokens each	32K tokens each
1	19.1 GB	20.6 GB
5	21.0 GB	28.1 GB
8	22.4 GB	33.7 GB
16	26.2 GB	48.7 GB
32	33.7 GB	78.8 GB
64	48.7 GB	139 GB

One card, with vLLM's small-card settings.

From the model card

What ByteDance says about UI-TARS-1.5

Read the model card

We shared the latest progress of the UI-TARS-1.5 model in our blog, which excels in playing games and performing GUI tasks.

Introduction

UI-TARS-1.5, an open-source multimodal agent built upon a powerful vision-language model. It is capable of effectively performing diverse tasks within virtual worlds.

Leveraging the foundational architecture introduced in our recent paper, UI-TARS-1.5 integrates advanced reasoning enabled by reinforcement learning. This allows the model to reason through its thoughts before taking action, significantly enhancing its performance and adaptability, particularly in inference-time scaling. Our new 1.5 version achieves state-of-the-art results across a variety of standard benchmarks, demonstrating strong reasoning capabilities and notable improvements over prior models.

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

Application: https://github.com/bytedance/UI-TARS-desktop

Performance

Online Benchmark Evaluation

Benchmark type	Benchmark	UI-TARS-1.5	OpenAI CUA	Claude 3.7	Previous SOTA
Computer Use	OSworld (100 steps)	42.5	36.4	28	38.1 (200 step)
	Windows Agent Arena (50 steps)	42.1	-	-	29.8
Browser Use	WebVoyager	84.8	87	84.1	87
	Online-Mind2web	75.8	71	62.9	71
Phone Use	Android World	64.2	-	-	59.5

Grounding Capability Evaluation

Benchmark	UI-TARS-1.5	OpenAI CUA	Claude 3.7	Previous SOTA
ScreensSpot-V2	94.2	87.9	87.6	91.6
ScreenSpotPro	61.6	23.4	27.7	43.6

Poki Game

Model	2048	energy	free-the-key	Gem-11	hex-frvr	Infinity-Loop	Maze:Path-of-Light	shapes	snake-solver	wood-blocks-3d	yarn-untangle	laser-maze-puzzle	tiles-master
OpenAI CUA	31.04	32.80	0.00	46.27	92.25	23.08	35.00	52.18	42.86	2.02	44.56	80.00	78.27
Claude 3.7	43.05	41.60	0.00	0.00	30.76	2.31	82.00	6.26	42.86	0.00	13.77	28.00	52.18
UI-TARS-1.5	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00

Minecraft

Task Type	Task Name	VPT	DreamerV3	Previous SOTA	UI-TARS-1.5 w/o Thought	UI-TARS-1.5 w/ Thought
Mine Blocks	(oak_log)	0.8	1.0	1.0	1.0	1.0
	(obsidian)	0.0	0.0	0.0	0.2	0.3
	(white_bed)	0.0	0.0	0.1	0.4	0.6
	200 Tasks Avg.	0.06	0.03	0.32	0.35	0.42
Kill Mobs	(mooshroom)	0.0	0.0	0.1	0.3	0.4
	(zombie)	0.4	0.1	0.6	0.7	0.9
	(chicken

Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.

Details

Published under	ByteDance-Seed
Type	Language models
Task	Vision + text
Runs with	transformers
Popularity	741k downloads / month
Weights	16.6 GB, UI-TARS-1.5-7B (FP32, file size)

Languages