Stable Audio 3 open weights: run them yourself or call a hosted API
Stable Audio 3 Small SFX, Small and Medium are open-weight; Large is on Stability's API. Sume's music route lists Lyria ids only, so weights mean self-hosting.
"Open weights" for Stable Audio 3 means you download the model and run it on your own hardware. Stability AI says Stable Audio 3.0 Small SFX, Small and Medium are open-weight on Hugging Face, and Large is available through the Stability API. Sume's Music Router lists no Stable Audio id, so weights are a self-hosting path, and a hosted call from Sume goes to Lyria.
The Stable Audio facts are from Stability's announcement, read 2026-09-29. The Sume facts are from the Music Router and Music 1.0 docs.
Which Stable Audio 3 models are open-weight?
Stability released the family on May 20, 2026. These are its statements, and we did not run any of the models.
| Model | How Stability says you get it |
|---|---|
| Small SFX | Open-weight, on Hugging Face |
| Small | Open-weight, on Hugging Face |
| Medium | Open-weight, on Hugging Face |
| Large | Through the Stability API |
What does the license say?
Stability names a Community License and an Enterprise License for organizations with $1M or more in revenue. It also says the models were trained on fully licensed data. Read the license text itself before you build on the weights; this post is not a license summary.
Which music models does Sume list?
The Music Router's routable ids are sume/music-auto (the default), lyria-3.5 and lyria-3-pro. An unknown id fails with 400 model_not_found and a catalog_url, so a Stable Audio id is not accepted. List the current catalog with GET /v1/music-router/models.
curl https://api.sume.com/v1/music-router/models \
-H "Authorization: Bearer $SUME_API_KEY"How do I choose between weights and a hosted route?
- Choose weights when you need to run the model on your own machines and can supply the hardware and the license review.
- Choose a hosted route when you want a job, a result URL and a fixed price per generation without running a model.
- Whichever you pick, generate a short test with your own prompt first; neither vendor's own claims replace listening.
What does a hosted call on Sume cost?
Every Music Router model charges the fixed Music price per audio generation, and Music 1.0 lists $0.125 per accepted generation, independent of prompt length. The catalog lists each model's provider list price for reference. Self-hosting has no per-call price from Sume, but you pay for the hardware and time to run the weights.
For Stable Audio 3 Large, Stability says it is available through the Stability API. Its request shape and price are on Stability's own pages, which this post does not cover.
What should I ask before self-hosting?
Open weights move the work to you. These are questions to answer first, none of them taken from Stability's page:
- Which GPU or CPU will run it, and how many tracks per hour do you need?
- Who reviews the license, and does your organization's revenue change which one applies?
- How will you version the weights so a track can be reproduced later?
Sources
Related posts
More in Models
- Stable Audio 3 max length: six minutes, and how to reach it on Sume
Stability says Stable Audio 3 Medium and Large pass six minutes. Sume's Lyria music runs up to a few minutes; join separate takes for more with Timeline audio.
- Text to speech API with emotion, speed and volume controls
Sume's text to speech API takes generation_config: speed 0.6 to 1.5, volume 0.5 to 2.0 and an optional emotion guide of up to 64 characters.
- Text to speech sample rate: which output_format to pick
Sume's TTS output_format takes mp3, wav or raw, six sample rates from 8000 to 48000 Hz and four PCM encodings. The default, and when to send wav.
- Veo 3 shutdown: which Gemini API models retired and what replaces them
Google retired veo-3.0-generate-001, veo-3.0-fast-generate-001 and veo-2.0-generate-001 on 2026-06-30. Its listed replacements are the Veo 3.1 preview models.
Written by Sume