Stability AI launches on-device music generator with open weights
Stability AI’s new audio models can run on-device and make tracks up to two minutes long, while its largest version can generate full songs of 6 minutes 20 seconds.
Stability AI has pushed its music tools into a new phase with Stable Audio 3.0, a four-model family that includes on-device generators and a larger system capable of full compositions lasting up to 6 minutes 20 seconds. The release matters because it moves generative AI further from short clips and toward production-ready music, while sharpening the debate over who benefits when machines can make usable tracks on phones, laptops and local hardware.
The company said the lineup includes small SFX, small, medium and large variants. The two small models each have 459 million parameters and are designed for on-device sound and music creation, with track lengths of up to two minutes. The medium model has 1.4 billion parameters, and the large model has 2.7 billion parameters; those versions are aimed at longer, fuller songs. Stability AI also said three of the four models ship as open weights, while the largest model is available through its API and enterprise or self-hosted deployment.
Stability AI is leaning hard on licensing as the industry tries to answer copyright criticism that has shadowed music AI from the start. The company said Stable Audio 3.0 was trained on fully licensed data, and that users own their outputs under the Community License or Enterprise License. That language is meant to reassure musicians, rights holders and labels that the company is building on cleaner ground than many earlier AI systems, which were often accused of scraping copyrighted material without permission.
The launch also builds on a fast product cadence. Stability Audio first arrived in September 2023 with commercially usable clips of up to 90 seconds, then expanded in April 2024 with Stable Audio 2.0, which added up to three-minute tracks and audio-to-audio editing. The new version widens the ceiling again, with variable-length generation and what Stability AI describes as full song composition on portable devices.
Still, the broader conflict has not disappeared. A later 2025 partnership between Universal Music Group and Stability AI pointed toward a more regulated, licensed future for AI music tools, but 2026 reports of copyright lawsuits involving Stability AI and AudioSparx showed how unsettled the legal landscape remains. For musicians and producers, Stable Audio 3.0 offers a faster and more flexible toolset; for rights holders, it is another test of whether the market can reward innovation without eroding control over the work that trained the next generation of machines.
Know something we missed? Have a correction or additional information?
Submit a Tip
