whatmodel.sh / image / HunyuanImage 3.0

HunyuanImage 3.0

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

HunyuanImage 3.0 is an open-weights image model from Tencent, released September 2025. Its published size is 80B parameters. Licence: tencent-hunyuan-community.

57.2±1.7#59 of 144 image modelsfrom 12 benchmark results across 9 sources

Score within the LMArena text-to-image comparison set only.

$ how it compares

The image models ranked either side of HunyuanImage 3.0, out of 197 with a score, then more from Tencent.

#modelscorewhereRAMreleased
66Qwen-Image-Edit-2511 Alibaba58±2api $0.030/image +local~37GB2025-12
67Recraft V4.1 Flash Recraft57±3api $0.0070/image2026-09
68HunyuanImage 3.0 Tencent57±2api $0.080/image +local~49GB2025-09
69Emu3.5 BAAI57±5local~85GB2025-10
70JoyAI-Image-Edit JD (JoyAI)57±3comparison only2026-03
more from Tencent
38HunyuanImage 3.0-Instruct Tencent62±3api $0.090/image +local~203GB2026-01
65HunyuanImage 2.1 Tencent58±2api $0.10/image +local~64GB2025-09

see all image models · find the ones that fit your machine

$ what it's good at

text to image▂▅▅▃▄▄█▃▁▂▁▁ #50 of 126 · 87% of its score · 10 tests

Beats Wan 2.7 Image, HunyuanImage 2.1, Recraft V4.1 Flash here, though they rank higher overall.

testresultvs the others in its rundifficultycountswhy
LMArena text-to-image1149.5 ✎▂▂▃▄▇█▄▂▄▂·▂ beats 44/75avg 1027▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 41%typical
GenExam (relaxed overall, GPT-5 judge)19.1% ✎▆▆·▂▂····▂▂█ beats 2/13hard · avg 20%▇▇▇▇▇▇ 15%sharp · at its level · noisy
UltraText Bench (Composite, bilingual dense text rendering)29.9% ✎▁▂▃█▃▁▂▃·▁·▁ beats 4/22hard · avg 28%▇▇▇▇▇ 11%sharp · at its level · noisy
FEPBench instruction faithfulness41.0% ✎▄·█···▄▄▄▄▄▄ beats 2/8medium · avg 36%▇▇ 5.4%blunt · at its level
WISE (WiScore overall, GPT-4o judge)57.0% ✎▄·▆▂▆█▄▆▃▄·▄ beats 18/22medium · avg 60%▇▇ 4.9%at its level
Qwen-Image-Bench (Q-Judger, v1)50.8% ✎█▅▅·▅▅█·█··▃ beats 4/17medium · avg 46%▇▇ 4.5%blunt · at its level
HPSv3 score of Cosmos 3 report text-to-image evaluation images11.9300 ✎▄···█··▄·▄·▄ beats 5/5avg 11.076▇ 2.5%blunt
ChineseWord (LongCat-Image 8,105-character set, PPOCRv5) overall49.3% ✎▄█·········▄ beats 0/3medium · avg 58%▇ 0.7%blunt · at its level · noisy
2 more tests
testresultvs the others in its rundifficultycountswhy
CVTG-102ch PNED (Chinese character rendering, Cosmos 3 report)71.3% ✎▄·······▄█▄▄ beats 3/5medium · avg 60%▇ 0.7%blunt · at its level · noisy
CVTG-500L PNED (English long-prompt text rendering, Cosmos 3 report)87.7% ✎█······█████ beats 3/5easy · avg 85%▇ 0.0%blunt
other tests2 not in a task · 12% of its score
testresultvs the others in its rundifficultycountswhy
Rapidata Benchmark.ai image benchmark (overall Elo)924.8 ✎▁▁▁▄▂█▂▁▁▂▁▁ beats 5/34avg 875▇▇▇ 6.4%typical
UniGenBench + Physical-AI prompts, All 1,170 (Cosmos 3 report)84.0% ✎▄···█··▄···█ beats 1/5easy · avg 78%▇▇ 5.8%typical

$ where to run it

Through an API

api · Replicate (PrunaAI optimized) · $0.080/image · replicate.com ↗ · price sourceUSD per output image on Replicate's tencent/hunyuan-image-3 model (billing: per output image; hardware time not added). Current version served from replicate-pruna/hunyuan-image-3 (PrunaAI optimized). Product: Replicate model API. API model: tencent/hunyuan-image-3. Settings: one output image. Price basis: published list rate. Currency: USD. Replicate official model page, observed 2026-09-25.

Other ways to run it

version

  1. Run it as published · needs ~192.5 GB, NVIDIA only · from github.com/Tencent-Hunyuan
    git clone https://github.com/Tencent-Hunyuan/HunyuanImage-3.0.git && cd HunyuanImage-3.0 && pip install torch==2.8.0 torchvision==0.23.0 torchaudio==2.8.0 --index-url https://download.pytorch.org/whl/cu128 && pip install -r requirements.txt && hf download tencent/HunyuanImage-3.0 --local-dir ./HunyuanImage-3 && python3 run_image_gen.py --model-id ./HunyuanImage-3 --verbose 1 --prompt "a cat holding a sign" --bot-task image --image-size "1024x1024" --save ./image.png --rewrite 0

Official weights and documentation: huggingface.co/tencent ↗

$ how the score was worked out · LMArena leaderboard dataset 41%, GenExam leaderboard (OpenGVLa… 15%, UltraText Bench: A Comprehens… 11% and 6 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
LMArena leaderboard dataset1▇▇▇▇▇▇▇▇▇▇▇▇ 41%CC BY 4.0
GenExam leaderboard (OpenGVLab/GenExam README, relaxed score)1▇▇▇▇▇ 15%MIT
UltraText Bench: A Comprehensive Bilingual Benchmark for Evaluating Visual Text Rendering in Image Generation (arXiv:2610.09823v1)1▇▇▇ 11%CC-BY-4.0
Cosmos 3: Omnimodal World Models for Physical AI (arXiv:2606.02800v4)4▇▇▇ 9%CC-BY-4.0
Rapidata/text-to-image-human-preference-6M (Benchmark.ai Image Benchmark) published dataset-card rankings1▇▇ 6%CC-BY-4.0
Faithful, Enriched, and Precise: Benchmarking Natural-Science Illustration Generation by T2I models (FEPBench)1▇▇ 5%CC0-1.0
WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation (arXiv v4)1▇ 5%CC-BY-SA-4.0
Qwen-Image-Bench: From Generation to Creation in Text-to-Image Evaluation1▇ 4%CC-BY-4.0
Meituan LongCat — LongCat-Image technical report1▇ 1%CC BY 4.0
prior · pull toward the middle▇ 1%
weighted average= 57.2

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json