Stable Audio 3

Stability AI's open instrumental music and sound-effect models: up to 6:20 of audio, from a CPU-friendly Small to an 8 GB-GPU Medium.

In short. Stable Audio 3 makes instrumental music and sound effects from a text prompt, up to six minutes long, and can also extend or edit existing audio. The Small models run on a laptop CPU. It does not sing, and its license is free for commercial use only below US$1 million in revenue.

Summary

Stability AI released Stable Audio 3 on 20 May 2026 as a family of four models, three of them with open weights: Small Music and Small SFX, which each make up to two minutes of audio, and Medium, which does both music and sound effects for up to 6 minutes 20 seconds. A Large model with 2.7 billion parameters is available only through Stability's API and enterprise self-hosting. All of them are latent diffusion transformers on a new semantic-acoustic autoencoder called SAME, trained with flow matching and then with adversarial post-training so that eight sampling steps are enough. They make 44.1 kHz stereo audio from a text prompt, can inpaint, continue or edit existing audio, and the repository supports LoRA fine-tuning.

The technical report (arXiv 2605.17991) says the models were trained on licensed audio from AudioSparx, about 806,000 recordings, plus about 470,000 Creative Commons recordings from Freesound, with the music in the Freesound portion screened for copyright. The models make instrumental music and sound effects, not songs with sung lyrics. Stability reports a peak of 5 to 6.5 GB of VRAM for Medium and under 2.5 GB for Small, so Medium fits a consumer GPU with 8 GB, and Small generates two minutes of audio in about 6 seconds on the CPU of a MacBook Pro M4. Stability says its models set a new state of the art among open-weight systems for instrumental music and sound effects, which is a claim from its own evaluation. Medium needs CUDA and Flash Attention 2.

The weights use the Stability AI Community License: free for research and non-commercial use, and for commercial use below US$1 million in annual revenue, with registration and a "Powered by Stability AI" credit. The license also bars using the models or their outputs to build another foundation model. The Hugging Face repositories are gated. In the 30 days to 11 November 2026, Hugging Face counted about 113,000 downloads of Medium, 64,000 of Small SFX and 21,000 of Small Music.

Good for

  • Drafts of background music, loops and sound effects on a laptop or an 8 GB GPU
  • Extending, inpainting or editing existing audio clips
  • Phone and laptop apps that need on-device generation, with the Small models
  • Studios that want a model trained on licensed and Creative Commons data

Watch out for

  • It makes instrumental music and sound effects only, with no sung lyrics
  • Free commercial use ends at US$1 million in annual revenue, and you must register with Stability AI
  • Medium needs a CUDA GPU and Flash Attention 2, and the repositories are gated
  • The state-of-the-art claim comes from Stability's own evaluation, and the 2.7B Large model is API-only

Performance claims come from the developer and have not been independently verified.

How much memory does it need?

Memory figures for each variant of Stable Audio 3. Image, video, speech and music pipelines also load a text encoder and a decoder, so the memory needed can be larger than the main checkpoint.

Variants Parameters Download size Memory needed Fits on
Small Music (up to 2 min) FP32 460M 3.5 GB 4.02 GB Laptop or phone
Small SFX (sound effects, up to 2 min) FP32 460M 3.5 GB 4.02 GB Laptop or phone
Medium (music and sound effects, up to 6 min 20 s) FP32 1.4B 11 GB 8 GB Laptop or phone

Download size covers every component needed to run the model, such as the generator, text encoder and decoder. Memory needed is a stated or published measured minimum where one exists; otherwise it is the download size plus about 15%. Larger images, longer clips and higher resolutions need more. The developer states a minimum of 8 GB of GPU memory.

The license in plain words

Stability AI's license for its open models. Research and non-commercial use is free, and so is commercial use while you and your affiliates earn under US$1 million a year, but commercial users must register with Stability AI and show "Powered by Stability AI". Above US$1 million the license ends and you must ask for an enterprise license. You own the outputs, to the extent the law allows.

  • Attribution or model name must be shown
  • Free commercial use only below a revenue cap
  • Acceptable-use policy applies

More about Stability Community

See also

Frequently asked questions

What is Stable Audio 3?
Stable Audio 3 makes instrumental music and sound effects from a text prompt, up to six minutes long, and can also extend or edit existing audio. The Small models run on a laptop CPU. It does not sing, and its license is free for commercial use only below US$1 million in revenue.
Can I use Stable Audio 3 commercially?
Yes, with conditions. Stable Audio 3 is released under the Stability Community, which allows commercial use but adds extra rules: Attribution or model name must be shown; Free commercial use only below a revenue cap; Acceptable-use policy applies. Read the license text before you ship a product.
Is Stable Audio 3 open source?
Stable Audio 3 is open-weight, not open source in the strict sense. You can download the model files, but the Stability Community adds conditions that the Open Source Initiative definition does not allow.
Can I run Stable Audio 3 on my own computer?
Yes. The smallest variant, Small Music (up to 2 min), needs about 4.02 GB of memory at 4-bit precision, which fits a modern laptop or a mid-range graphics card.
When was Stable Audio 3 released?
Stable Audio 3 was released by Stability AI on May 20, 2026.

Official links

Sources

Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.

Release date and facts last verified November 11, 2026.

Related models

Model Released Parameters Context License Smallest variant needs
Magenta RealTime 2 Google DeepMind 230M – 2.4B – CC BY 4.0 (Permissive) 2.53 GB Laptop or phone
ACE-Step 1.5 ACE-Step 2B – 4B – MIT (Permissive) 4 GB Laptop or phone
MOSS-SoundEffect v2.0 OpenMOSS 1.3B – Apache 2.0 (Permissive) 13 GB Laptop or phone
MiniMax Music 3 MiniMax 11.1B – MiniMax Music3 Community (Conditional) 24 GB Gaming GPU