whatmodel.sh / image / Stable Diffusion 3.5 Large
Stable Diffusion 3.5 Large
↗ share
+ contribute
Stable Diffusion 3.5 Large is an open-weights image model from Stability AI, released October 2024. Its published size is 8.1B parameters. Licence: stabilityai-community.
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of Stable Diffusion 3.5 Large, out of 197 with a score, then more from Stability AI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 135 |  FLUX.1 [schnell] | 42±2 | api $0.0030/image +local | ||
| 136 |  OmniGen2 | 42±2 | api $0.15/image +local | ||
| 137 | Stable Diffusion 3.5 Large | 42±1 | api $0.065/image +local | ||
| 138 |  Show-o2 7B | 42±3 | local | ||
| 139 |  Playground v2 (1024px aesthetic) | 41±4 | local | ||
| more from Stability AI | |||||
| 141 | Stable Diffusion 3.5 Medium | 41±3 | api $0.035/image +local | ||
| 148 | Stable Diffusion XL 0.9 (base) | 40±6 | local | ||
| 149 | Stable Diffusion 3 Medium | 40±2 | api $0.035/image +local | ||
$ what it's good at
text to image▂▅▅▃▄▄█▃▁▂▁▁ #86 of 126 · 88% of its score · 8 tests
Beats Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M), VQ-Diffusion (ITHQ), FLUX.1 Kontext [dev] here, though they rank higher overall. Trails Playground v2.5, Playground v2 (1024px aesthetic), Show-o2 7B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LMArena text-to-image | 938.3 ✎ | ▃▂▃▄▇█▄▂▄▂·▂ beats 1/75 | 27% |
| GenExam (relaxed overall, GPT-5 judge) | 15.9% ✎ | ▄█▃▂▂▂▆▄·▂·▂ beats 6/16 | 19% |
| T2I-ReasonBench (overall) | 55.2% ✎ | ▅▂█▂▂▂····▂▂ beats 7/13 | 16% |
| GenEval | 59.1% ✎ | ██·▄·▄·█···▄ beats 1/8 | 10% |
| ServImage task acceptance (human payment decisions, commercial design tasks) | 8.6% ✎ | ▂▃··▂··▃█··▂ beats 1/11 | 4.9% |
| MMMG image generation (average over 15 tasks) | 33.2% ✎ | █··█▄▄·▄▄·▄▄ beats 0/9 | 4.6% |
| WISE (WiScore overall, GPT-4o judge) | 46.0% ✎ | ▄·▆▂▆█▄▆▂▄·▄ beats 11/22 | 3.8% |
| UltraText Bench (Composite, bilingual dense text rendering) | 7.5% ✎ | ▃▂▂█▃▁▂▃·▁·▁ beats 0/22 | 3.4% |
image to image / editing▁▁·▃▂▃█▃▃▄▁▁ #57 of 62 * · 4.9% of its score · 1 test
Beats UNO here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| ServImage task acceptance (human payment decisions, commercial design tasks) | 8.6% ✎ | ▂▃··▂··▃█··▂ beats 1/11 | 4.9% |
other tests3 not in a task · 11% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| VVRBench accuracy (10,000 verifiable synthetic-scene tasks) | 2.5% ✎ | ▃·█▅▃·▅▃···▃ beats 1/10 | 5.9% |
| Rapidata Benchmark.ai image benchmark (overall Elo) | 794.4 ✎ | ▃▁▁▄▂█▂▁▁▂▁▁ beats 0/34 | 4.2% |
| Knowledge Visualization T2I (Brief Caption, English overall) | 31.1% ✎ | ▅·▃·▃··▃·█▃▃ beats 4/9 | 0.7% |
$ where to run it
Through an API
api · Replicate · $0.065/image · replicate.com ↗ · price sourceUSD per output image on Replicate's stability-ai/stable-diffusion-3.5-large model (billing: per output image; hardware time not added). Product: Replicate model API. API model: stability-ai/stable-diffusion-3.5-large. Settings: one output image. Price basis: published list rate. Currency: USD. Not a Stability AI direct API tariff. Replicate official model page, observed 2026-09-25.
Other ways to run it
version
- Run it as published · needs ~19.94 GB, NVIDIA only · from stability.ai (official)
uv run --with diffusers --with transformers --with accelerate --with torch --with sentencepiece --with protobuf python -c "import torch; from diffusers import StableDiffusion3Pipeline; p=StableDiffusion3Pipeline.from_pretrained('stabilityai/stable-diffusion-3.5-large', torch_dtype=torch.bfloat16).to('cuda'); p('a cat holding a sign', num_inference_steps=28, guidance_scale=3.5).images[0].save('out.png')"
Official weights and documentation: huggingface.co/stabilityai ↗
$ how the score was worked out · LMArena leaderboard dataset 27%, GenExam: A Multidisciplinary… 19%, T2I-ReasonBench: Benchmarking… 16% and 7 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
$ improve this page
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