whatmodel.sh / image / Seedream 4.0

Seedream 4.0

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

Seedream 4.0 is a closed-weights image model from ByteDance Seed, released September 2025. Licence: proprietary.

61.0±1.2#38 of 144 image modelsfrom 14 benchmark results across 11 sources

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

$ how it compares

The image models ranked either side of Seedream 4.0, out of 197 with a score, then more from ByteDance Seed.

#modelscorewhereRAMreleased
43Wan 2.6 Text-to-Image Alibaba61±3api $0.030/image2026-01
44Reve V1.1 Reve61±4comparison only
45Seedream 4.0 ByteDance Seed61±1api $0.040/image2025-09
46Recraft V4.1 Pro Recraft61±3api $0.21/image2026-05
47ERNIE-Image-Turbo Baidu61±3api $0.010/image +local~17GB2026-04
more from ByteDance Seed
12Seedream 5.0 Pro ByteDance Seed75±2api $0.050/image2026-07
29Seedream 5.0 Lite ByteDance Seed65±2api $0.035/image2026-02
73Seedream 3.0 ByteDance Seed56±2comparison only2025-04

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

$ what it's good at

text to image▂▅▅▃▄▄█▃▁▂▁▁ #36 of 126 · 68% of its score · 5 tests

Beats Wan 2.7 Image Pro, Wan 2.7 Image here, though they rank higher overall. Trails ERNIE-Image, ERNIE-Image-Turbo here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
GenExam (relaxed overall, GPT-5 judge)53.2% ✎▄█▂▂▂▂▆▄·▂·▂ beats 11/16hard · avg 26%▇▇▇▇▇▇▇▇▇▇▇▇▇ 34%sharp · at its level
LMArena text-to-image1139.9 ✎▂▂▃▄▇█▄▂▄▂·▂ beats 40/75avg 1027▇▇▇▇▇▇▇▇▇ 22%typical
ServImage task acceptance (human payment decisions, commercial design tasks)31.1% ✎▂▃··▂··▃█··▂ beats 8/11hard · avg 18%▇▇▇▇ 10%sharp · at its level
Qwen-Image-Bench (Q-Judger, v1)56.2% ✎█▅▅·▅▅█·█··▃ beats 10/17medium · avg 46%▇ 2.3%blunt · at its level
ChineseWord (LongCat-Image 8,105-character set, PPOCRv5) overall58.5% ✎▄█·········▄ beats 2/3medium · avg 58%▇ 0.4%blunt · at its level · noisy
image to image / editing▁▁·▃▂▃█▃▃▄▁▁ #30 of 62 · 36% of its score · 7 tests

Beats FLUX.2 [flex], GPT Image 1 here, though they rank higher overall. Trails Kling Image O1 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
ServImage task acceptance (human payment decisions, commercial design tasks)31.1% ✎▂▃··▂··▃█··▂ beats 8/11hard · avg 18%▇▇▇▇ 10%sharp · at its level
Arena image edit1270.2 ✎▂▁▁▂▆▇█▄▆▅▁▂ beats 25/53avg 1153▇▇▇ 8.2%typical
CEdit-Bench-CN overall (G_O, GPT-4o judge)75.7% ✎▃·········█▂ beats 5/7medium · avg 42%▇▇ 5.5%sharp · noisy
GEBench GUI state generation (English GE-Score)46.5% ✎▄▄····▄▄█▄·█ beats 2/8hard · avg 34%▇▇ 4.1%at its level
CEdit-Bench-EN overall (G_O, GPT-4o judge)75.8% ✎▄·▄···▄██··▄ beats 5/7easy · avg 66%▇ 3.7%typical
UnicBench (EN, overall)80.4% ✎▂····▂▄▂▄·█▂ beats 10/11easy · avg 73%▇ 3.0%typical
CREval-Bench creative image editing (Overall, GPT-4o QA evaluator)83.4% ✎▄█▄··█··▄·▄▄ beats 8/8easy · avg 74%▇ 2.0%blunt
other tests3 not in a task · 5.1% of its score
testresultvs the others in its rundifficultycountswhy
EdiVal multi-turn editing overall (EdiVal-O after turn 3)59.8% ✎▂·▂▃·▂·▃▂·▃█ beats 14/14medium · avg 42%▇ 3.0%blunt · at its level
ViGoR-Bench zero-shot visual reasoning, image models (Result Avg)41.5% ✎▂···▂·▂█▆▅·▂ beats 12/15hard · avg 31%▇ 1.3%at its level
Knowledge Visualization T2I (Brief Caption, English overall)39.6% ✎▅·▃·▃··▃·█▃▃ beats 9/9hard · avg 28%▇ 0.7%blunt · at its level

$ where to run it

Through an API

api · DeepInfra · $0.040/image · deepinfra.com ↗ · price sourceUSD per generated image on DeepInfra; the page lists one flat per-image rate (default size 2K; 1K/4K and explicit sizes accepted). Product: DeepInfra text-to-image. API model: ByteDance/Seedream-4. Settings: n=1, text-to-image. Price basis: published list rate. Currency: USD. Not the BytePlus ModelArk direct tariff. DeepInfra official model page, observed 2026-09-25.

$ how the score was worked out · GenExam: A Multidisciplinary… 34%, LMArena leaderboard dataset 30%, ServImage: An Image Generatio… 10% and 8 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
GenExam: A Multidisciplinary Text-to-Image Exam (arXiv v5)1▇▇▇▇▇▇▇▇▇▇ 34%CC-BY-4.0
LMArena leaderboard dataset2▇▇▇▇▇▇▇▇▇ 30%CC BY 4.0
ServImage: An Image Generation and Editing Benchmark from Real-world Commercial Imaging Services (arXiv:2604.24023v2)1▇▇▇ 10%CC-BY-4.0
Meituan LongCat — LongCat-Image technical report3▇▇▇ 10%CC BY 4.0
GEBench: Benchmarking Image Generation Models as GUI Environments1▇ 4%CC-BY-4.0
UnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched Edits1▇ 3%CC-BY-4.0
EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing (arXiv:2509.13399v3)1▇ 3%CC-BY-4.0
Qwen-Image-Bench: From Generation to Creation in Text-to-Image Evaluation1▇ 2%CC-BY-4.0
CREval: An Automated Interpretable Evaluation for Creative Image Manipulation under Complex Instructions (arXiv:2603.26174v1)1▇ 2%CC-BY-4.0
ViGoR-Bench: How Far Are Visual Generative Models From Zero-Shot Visual Reasoners? (arXiv:2603.25823v1)1▇ 1%CC-BY-4.0
Knowledge Visualization: A Benchmark and Method for Knowledge-Intensive Text-to-Image Generation (arXiv:2604.22302v1)1▇ 1%CC-BY-4.0
prior · pull toward the middle▇ 1%
weighted average= 61.0

$ 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