Model Follower.

Up to date

Rebuilt
2026-10-25 06:54 UTC
Release radar
117 on watch
Hub snapshot
2026-10-25
Re-checked
daily · every number sourced and dated

Models › UI-TARS-1.5 family

UI-TARS-1.5-7B

Hugging Face ↗ Follow changes

ByteDance-Seed Apache 2.0 out 18 months ago

8.29B parameters, images and text in, text out, 125K tokens of context. 2 providers serve it from $0.10 per million input tokens.

Verified 2026-10-25 Rank #122 in LLMs · score 95.8 method v1 Sources 5 Downloads 30d 118.4k · #299 of 619 tracked

Can I run it?5.0GB at INT4Estimatefits a 16 GB cardmethod
What does it cost?$0.10/1M incheapest of 1Parasailbf162026-10-24
Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-24
Do people use it?118.4kdownloads, 30 days#299 of 619 we track · top 48%+2 likes, downloads −51% this week613 likesHugging Face2026-10-24

Worth reading

Picked by rule from 2 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Where to run it

Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.

#Cheapest 2In /1MOut /1MContextQuantSeen
1 Parasail $0.10 $0.20 125K bf16 Oct 24
2 Featherless — — — — Oct 25

All 2 providers →

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

Served by 2 API providers; 13 quantized builds for local use.

  • As a service2Featherless, Parasail
  • On own hardware13quantized builds
  • Built on1512 fine-tunes · 2 adapters · 1 merges
  • In apps15Hugging Face Spaces

Top Spaces running it: prithivMLmods/CUA-GUI-Operator ♥ 12 · Mungert/UI-TARS-1.5-7B ♥ 5 · Stremly/uitars ♥ 2 · omar0scarf/ui-tars-api ♥ 1

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 16 GB card at INT4

About 5.0 GB at INT4, 10 GB at FP8, 20 GB at BF16: weights only, so leave room for the KV cache.

16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 5.0 GB of weights (estimate; KV cache and activations extra)INT45.0 GBFP8: about 10 GB of weights (estimate; KV cache and activations extra)FP810 GBBF16: about 20 GB of weights (estimate; KV cache and activations extra)BF1620 GB
16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 5.0 GB of weights (estimate; KV cache and activations extra)INT45.0 GBFP8: about 10 GB of weights (estimate; KV cache and activations extra)FP810 GBBF16: about 20 GB of weights (estimate; KV cache and activations extra)BF1620 GB

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

2 dated articles and 11 repo commits so far

Teal is the weights (2025-04-16), amber triangles are 2 dated articles (each links to it), grey ticks are repo commits.

MayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2025-04-18: 2 repo commits2025-04-17: 2 repo commits2025-04-16: 7 repo commitsUI-TARS-1.5: open-source computer use vision-language model · 2025-04-17UI-TARS-1.5-7B: a vision AI agent that reached SOTA in GUI grounding | lilting channel · 2026-02-05Weights 2025-04-16 · 2025-04-16Weights 2025-04-16now · 2 providers
MayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2025-04-18: 2 repo commits2025-04-17: 2 repo commits2025-04-16: 7 repo commitsUI-TARS-1.5: open-source computer use vision-language model · 2025-04-17UI-TARS-1.5-7B: a vision AI agent that reached SOTA in GUI grounding | lilting channel · 2026-02-05Weights 2025-04-16 · 2025-04-16Weights 2025-04-16now · 2 providers

What people built on it

28 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

28 derivatives on the Hub, led by 13 quantized

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

13 quantized (Hugging Face model tree)12 fine-tunes (Hugging Face model tree)1 merge (Hugging Face model tree)2 adapters (Hugging Face model tree)13 quantized12 fine-tunes1 merge2 adapters
13 quantized (Hugging Face model tree)12 fine-tunes (Hugging Face model tree)1 merge (Hugging Face model tree)2 adapters (Hugging Face model tree)13 quantized12 fine-tunes1 merge2 adapters
GGUF mradermacher/UI-TARS-1.5-7B-GGUF for llama.cpp, Ollama, LM Studio 630.6k downloads MLX mlx-community/UI-TARS-1.5-7B-8bit for Apple silicon 737 downloads GPTQ yujiepan/ui-tars-1.5-7B-GPTQ-W4A16g128 for vLLM, SGLang 179.7k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

26 rows

Benchmarks 26 rows

measured third-party leaderboard rowauthor-reported number from the model card or eval files (26)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.
BenchmarkValueSetting / notesReported byDate
author-reported 200 Tasks Avg.0.42UI-TARS-1.5 w/ Thoughtmodel author2026-10-07
author-reported 2048100UI-TARS-1.5model author2026-10-07
author-reported Android World64.2UI-TARS-1.5model author2026-10-07
author-reported Gem-11100UI-TARS-1.5model author2026-10-07
author-reported Infinity-Loop100UI-TARS-1.5model author2026-10-07
author-reported Maze:Path-of-Light100UI-TARS-1.5model author2026-10-07
author-reported Online-Mind2web75.8UI-TARS-1.5model author2026-10-07
author-reported ScreenSpotPro61.6UI-TARS-1.5model author2026-10-07
author-reported ScreensSpot-V294.2UI-TARS-1.5model author2026-10-07
author-reported WebVoyager84.8UI-TARS-1.5model author2026-10-07
author-reported chicken0.6UI-TARS-1.5 w/ Thoughtmodel author2026-10-07
author-reported cubinko0UI-TARS-1.5model author2026-10-07
author-reported energy100UI-TARS-1.5model author2026-10-07
author-reported free-the-key100UI-TARS-1.5model author2026-10-07
author-reported hex-frvr100UI-TARS-1.5model author2026-10-07
author-reported laser-maze-puzzle100UI-TARS-1.5model author2026-10-07
author-reported mooshroom0.4UI-TARS-1.5 w/ Thoughtmodel author2026-10-07
author-reported oak_log1UI-TARS-1.5 w/ Thoughtmodel author2026-10-07
author-reported obsidian0.3UI-TARS-1.5 w/ Thoughtmodel author2026-10-07
author-reported shapes100UI-TARS-1.5model author2026-10-07
author-reported snake-solver100UI-TARS-1.5model author2026-10-07
author-reported tiles-master100UI-TARS-1.5model author2026-10-07
author-reported white_bed0.6UI-TARS-1.5 w/ Thoughtmodel author2026-10-07
author-reported wood-blocks-3d100UI-TARS-1.5model author2026-10-07
author-reported yarn-untangle100UI-TARS-1.5model author2026-10-07
Show all 26 rows (1 more)
BenchmarkValueSetting / notesReported byDate
author-reported zombie0.9UI-TARS-1.5 w/ Thoughtmodel author2026-10-07

What the card says

1 cited sentence

Summary generated, cited

UI-TARS-1.5-7B is an 8.29B parameter multimodal model designed for performing GUI tasks and playing games.[1][2]

  1. parameters human
  2. “excels in playing games and performing GUI tasks”

Generated 2026-10-07 15:01 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

4 use cases

What it is good at 4 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

21 fields

Listed in LLMs

Facts

Pipeline tagimage-text-to-text
Librarytransformers
ArchitecturesQwen2_5_VLForConditionalGeneration
Model typeqwen2_5_vl
Dtypebfloat16
Params by dtypeF32: 8,292,166,656
Size bandM
MoEno
Context length128,000 (config.json max_position_embeddings)
PapersarXiv:2501.12326, arXiv:2404.07972, arXiv:2409.08264, arXiv:2401.13919, arXiv:2504.01382, arXiv:2405.14573, arXiv:2410.23218, arXiv:2504.07981
LicenseApache 2.0 (apache-2.0)
Hub snapshot2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · Qwen2_5_VLForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen2_5_VLForConditionalGeneration · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for UI-TARS-1.5-7B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)20 GB · Single GPU (<24 GB)
FP8 (1 byte/param)10 GB · Single GPU (<24 GB)
INT4 (0.5 bytes/param)5.0 GB · Single GPU (<24 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

28 variants

Lineage

Original model Family: ByteDance-Seed/UI-TARS-1.5 28 derivatives on the Hub

quantized: 13finetune: 12merge: 1adapter: 2

Top variants (28 shown, ranked by downloads)

Relation
Quant
28 rows
Variant Relation Quant Params Band License Author Likes Downloads
mradermacher/UI-TARS-1.5-7B-GGUF quantized GGUF — — — mradermacher 29 630.6k
yujiepan/ui-tars-1.5-7B-GPTQ-W4A16g128 quantized GPTQ 8.34B M — yujiepan 1 179.7k
flin775/UI-TARS-1.5-7B-AWQ quantized AWQ 8.29B M apache-2.0 flin775 0 28.4k
mradermacher/UI-TARS-1.5-7B-i1-GGUF quantized GGUF — — — mradermacher 5 8k
Lucy-in-the-Sky/UI-TARS-1.5-7B-Q4_K_M-GGUF quantized GGUF — — apache-2.0 Lucy-in-the-Sky 1 3.3k
adriabama06/UI-TARS-1.5-7B-GGUF quantized GGUF — — apache-2.0 adriabama06 5 2.8k
adriabama06/UI-TARS-1.5-7B-Q4_K_M-GGUF quantized GGUF — — apache-2.0 adriabama06 1 2.9k
tanvirb/websight-7B finetune — 8.29B M apache-2.0 tanvirb 0 1.1k
yasserrmd/Fara-TARS-7B merge — 8.29B M apache-2.0 yasserrmd 5 712
Lucy-in-the-Sky/UI-TARS-1.5-7B-Q6_K-GGUF quantized Q6_K — — apache-2.0 Lucy-in-the-Sky 0 917
Lucy-in-the-Sky/UI-TARS-1.5-7B-Q8_0-GGUF quantized Q8_0 — — apache-2.0 Lucy-in-the-Sky 1 781
mlx-community/UI-TARS-1.5-7B-8bit quantized -8bit 8.29B M apache-2.0 mlx-community 0 737
what2up/UI-TARS-1.5-7B-Q4_K_M-GGUF quantized GGUF — — apache-2.0 what2up 1 591
TMLR-Group-HF/AgentHijack-Agent finetune — 8.29B M apache-2.0 TMLR-Group-HF 1 251
rosethelocalfem/UI-TARS-1.5-7B-Q4_K_M-GGUF quantized GGUF — — apache-2.0 rosethelocalfem 0 211
PengxiangLi/dart-gui-7b finetune — 8.29B M apache-2.0 PengxiangLi 1 106
aylinakkus/ui-tars-1.5-7b-idm-full-sft finetune — 8.29B M apache-2.0 aylinakkus 0 142
mlfoundations-cua-dev/ui-tars-1.5-7b-idm-full-sft-8-frames finetune — 8.29B M apache-2.0 mlfoundations-cua-dev 0 138
rremani1990/websight-7B finetune — 8.29B M apache-2.0 rremani1990 0 80
flin775/UI-Tars-1.5-7B-4bit-mlx quantized -4bit — — apache-2.0 flin775 0 73
aylinakkus/ui-tars-1.5-7b-idm-full-sft-diff finetune — 8.29B M apache-2.0 aylinakkus 0 55
adriabama06/UI-TARS-1.5-7B-exl2 finetune exl2 — — apache-2.0 adriabama06 1 4
XinShu3047/SKC-checkpoints finetune — — — — XinShu3047 1 0
aylinakkus/ui-tars-1.5-7b-idm-full-sft-2-frames finetune — 8.29B M apache-2.0 aylinakkus 0 42
mlfoundations-cua-dev/ui_tars_7b_easyr1_10k_hard_qwen7b_easy_gta1-4MP finetune — 8.29B M other mlfoundations-cua-dev 0 37
su-collaborations/ui-tars-model-webclick-all adapter — — — apache-2.0 su-collaborations 0 31
figai/UI-TARS-1.5-7B-GUI-Perturbed adapter — 8.29B M apache-2.0 figai 0 0
dhananjay8896/gemma4 finetune — — — apache-2.0 dhananjay8896 0 0

Repository changelog

11 commits

Repository changelog 11 commits

DateCommitSHA
2025-04-18 Upload folder using huggingface_hub 683d002
2025-04-18 Upload folder using huggingface_hub d9ad05f
2025-04-17 Upload folder using huggingface_hub b39d18f
2025-04-17 Upload folder using huggingface_hub 9e004e3
2025-04-16 Upload folder using huggingface_hub 3ed9359
2025-04-16 Upload folder using huggingface_hub 418fc43
2025-04-16 Upload folder using huggingface_hub f13a85c
2025-04-16 Upload folder using huggingface_hub d86c0d9
2025-04-16 Upload folder using huggingface_hub 1602243
2025-04-16 Upload folder using huggingface_hub 7a0efb1
2025-04-16 initial commit ba8f9f6

All providers

2 providers · 2 endpoints

Where to run 2 providers · 2 endpoints

Hosted endpoints observed for UI-TARS-1.5-7B, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting. Cheapest observed input price: $0.10/M.

ProviderModel refContextQuant$/M in$/M outtok/sTTFTToolsStatusObserved
Parasail
parasail, openrouter
 parasail-ui-tars-1p5-7b 125K bf16 $0.10 $0.20 — — — serverless 2026-10-24
Featherless
hf
 ByteDance-Seed/UI-TARS-1.5-7B — — — — — — — live 2026-10-25

Articles and coverage

2 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Analysis2

UI-TARS-1.5: open-source computer use vision-language model
seed-tars.com2025-04-17about the UI TARS 1.5 familyHN 4 pts · 1 comments3,109 words
UI-TARS-1.5-7B: a vision AI agent that reached SOTA in GUI grounding | lilting channel
lilting.ch2026-02-05710 words
ByteDance's Seed team released UI-TARS-1.5, a vision-language model based on Qwen2.5-VL that uses reinforcement learning and inference-time scaling for GUI agents. The lightweight 7B model reaches 49.6% on ScreenSpotPro, while the largest variant achieves 61.6% on that benchmark and 42.5% on OSWorld. A desktop application supports local execution on Windows, macOS, and browsers.written, reviewed, cited
  • The model outputs absolute coordinates, requiring conversion to the actual screen size for correct execution.
  • UI-TARS-1.5-7B achieves a 27.5% score on OSWorld, placing it in the same ballpark as Claude 3.7.
  • Vision AI approaches have limits on dynamic UIs and states that cannot be understood from screenshots alone.

Papers

8 papers

Papers 8

arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).

Similar models

3 models

Similar models

  • Qwen3-VL-8B-Instruct · 8.77B
    same size band (M), same task (image-text-to-text), more providers, permissive license
  • Qwen2.5-VL-7B-Instruct · 8.29B
    same size band (M), same task (image-text-to-text), fewer providers, permissive license
  • Qwen3.5-9B · 9.65B
    same task (image-text-to-text), more providers, permissive license

Sources

5 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:34 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:01 UTC · Last verified 2026-10-25

Follow UI-TARS-1.5-7B

Watch this model

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.