whatmodel.sh / image / Stable Diffusion 3.5 Medium
Stable Diffusion 3.5 Medium
↗ share
+ contribute
Stable Diffusion 3.5 Medium is an open-weights image model from Stability AI, released October 2024. Its published size is 2.5B parameters. Licence: stabilityai-community.
Model information sources
Release date ↗ · Weights licence ↗ · Parameter count ↗
2.5B = the MMDiT-X diffusion transformer as Stability states it; text encoders (CLIP-L, CLIP-G, T5-XXL) and VAE not included.
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of Stable Diffusion 3.5 Medium, out of 197 with a score, then more from Stability AI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 139 |  Playground v2 (1024px aesthetic) | 41±4 | local | ||
| 140 | Playground v2.5 | 41±2 | local | ||
| 141 | Stable Diffusion 3.5 Medium | 41±3 | api $0.035/image +local | ||
| 142 |  CogView4-6B | 41±6 | comparison only | ||
| 143 |  Midjourney V7 | 40±3 | comparison only | ||
| more from Stability AI | |||||
| 137 | Stable Diffusion 3.5 Large | 42±1 | api $0.065/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▂▅▅▃▄▄█▃▁▂▁▁ #89 of 126 * · 76% of its score · 2 tests
Beats Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M), VQ-Diffusion (ITHQ), OmniGen2 here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| T2I-ReasonBench (overall) | 52.2% ✎ | ▅▂█▂▂▂····▂▂ beats 5/13 | 62% |
| WISE (WiScore overall, GPT-4o judge) | 45.0% ✎ | ▄·▆▂▆█▄▆▂▄·▄ beats 10/22 | 15% |
other tests1 not in a task · 21% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| VVRBench accuracy (10,000 verifiable synthetic-scene tasks) | 2.8% ✎ | ▃·█▅▃·▅▃···▃ beats 2/10 | 21% |
$ where to run it
Through an API
api · $0.035/image · platform.stability.ai ↗ · price sourceUSD per generated image on the Stability AI Developer Platform API: the pricing table lists Stable Diffusion 3.5 Medium at 3.5 credits and states 'API usage is based on credits. 1 credit = $0.01' (3.5 x $0.01 = $0.035). Product: Stability AI REST API (v2beta stable-image/generate/sd3). API model: sd3.5-medium. Settings: text-to-image, one image per request. Price basis: credits x Stability's own USD per credit. The pricing page is client-rendered; the table (column 'Price (credits)') is in the page's own script bundle. SD 3.5 Large 6.5 credits, Large Turbo 4, Flash 2.5. Replicate's official stability-ai/stable-diffusion-3.5-medium lists the same $0.035 per output image. Currency: USD. Stability AI official pricing page, observed 2026-09-26.
Other ways to run it
- Run it as published · needs ~20 GB, NVIDIA only · from huggingface.co/stabilityai (official)
uv run --with torch --with diffusers --with transformers --with accelerate --with sentencepiece --with protobuf python -c "import torch; from diffusers import StableDiffusion3Pipeline; p=StableDiffusion3Pipeline.from_pretrained('stabilityai/stable-diffusion-3.5-medium', torch_dtype=torch.bfloat16).to('cuda'); p('A capybara holding a sign that reads Hello World', num_inference_steps=40, guidance_scale=4.5).images[0].save('out.png')"
Official weights and documentation: huggingface.co/stabilityai ↗
$ how the score was worked out · T2I-ReasonBench: Benchmarking… 62%, Verifiable Visual Rewards Tra… 21%, WISE: A World Knowledge-Infor… 15%
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
| source | results | share of the score |
|---|---|---|
 T2I-ReasonBench: Benchmarking Reasoning-Informed Text-to-Image Generation (Findings of ACL 2026) | 1 | 62% |
 Verifiable Visual Rewards Transfer from Synthetic Scenes to Natural Prompts (VVRBench), arXiv:2609.35641v1 | 1 | 21% |
| WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation (arXiv v4) | 1 | 15% |
| prior · pull toward the middle | 3% | |
| weighted average | = 40.7 |
$ 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