Stable Audio 3.0 Medium
Audio modelOpen weightsCard summarizedMeet Stable Audio 3.0, the model family built for artistic experimentation with open-weight models
Source: Stable Audio 3 Medium model card (Hugging Face) and Stable Audio 3 paper (arXiv 2605.17991)
Where to call it
- Self-host (weights)
stabilityai/stable-audio-3-medium - stable-audio-3 / stable-audio-tools
medium
Good for
- Music and sound effects generation of variable length
- Targeted editing of audio through inpainting
- Continuing short recordings
Overview
Stable Audio 3 M
Meet Stable Audio 3.0, the model family built for artistic experimentation with open-weight models
- Max length
- 6 min 20 s
- Languages
- —
- Modes
- text-to-audio (music and sound effects), audio inpainting
- Released
- 20 May 2026
Stable Audio 3.0 Medium is the larger of the two open-weights sizes in Stability AI's Stable Audio 3 family, released on 20 May 2026. It is a 1.4-billion-parameter latent diffusion transformer that generates 44.1 kHz stereo music and sound effects up to 6 min 20 s, with inpainting and continuation of existing recordings. Stability says it runs on consumer hardware: a few seconds on a MacBook Pro M4 and under 2 s on an H200. It was trained on licensed and Creative Commons audio. model card; paper; launch post
stabilityai/stable-audio-3-medium- Typical price
- —
- Max length
- 6 min 20 s
- Weights
- Open · Stability AI Community License (free under $1M annual revenue; Enterprise License above); bundled T5Gemma under Gemma Terms of Use
- Released
- 20 May 2026
- text-to-audio (music and sound effects)
- audio inpainting
- continuation
- LoRA fine-tuning