Why this is noise: Stability Audio 3.0 is a meaningful technical step up in open-weight audio generation, but the story here is licensing strategy, not model capability. The race to control AI music training data is now the defining competitive variable in this market.
Key Takeaways
- Stability AI has released Stable Audio 3.0, a family of four models ranging from 459M to 2.7B parameters, with the top two capable of generating full compositions of 6 minutes 20 seconds — more than double the 2024 version’s ceiling
- Three models are open-weight (small SFX, small, medium); the large model is API and self-hosted only, with an enterprise license required for companies earning over $1 million in revenue
- All models are trained on fully licensed data, backed by strategic partnerships with both Universal Music Group (October 2025) and Warner Music Group (November 2025)
- Stability AI is building a professional musicians suite and has hired Ethan Kaplan, former Chief Digital Officer at Universal Audio and Fender, to lead it
Stability AI has released Stable Audio 3.0, its most capable audio generation family to date, with its largest model producing professional-grade compositions over six minutes long and all models built on fully licensed training data.
The release marks a significant jump from Stable Audio 2.0, which topped out at around three minutes, and a major leap from the open-weight Stable Audio Open released in 2024, which was limited to 47 seconds. The new family spans four models: two small models at 459M parameters suited for on-device sound and music generation up to two minutes, a medium model at 1.4B parameters, and a large model at 2.7B parameters, with the latter two both capable of the full 6-minute-20-second output while maintaining musical structure and melodic coherence.
The licensing architecture is as important as the model capability. Stability AI’s partnerships with Universal Music Group and Warner Music Group, both signed in late 2025, give it a legally defensible training data position at a moment when the rest of the AI music industry is navigating active litigation. Suno is still fighting UMG and Sony Music in court after Warner Music settled and exited the joint lawsuit in November 2025, and the terms of that settlement remain under a discovery dispute that UMG and Sony are still trying to crack open.
“WMG is a leader in the global music and entertainment landscape, and we’re proud to partner with a company that shares our Artist First ethos.” — Prem Akkaraju, CEO, Stability AI
That legal backdrop makes Stability AI’s fully licensed positioning a structural advantage, not just a marketing claim. While Suno and Udio built audience and capability first and are now managing the legal fallout, Stability AI built the licensing infrastructure first and is now scaling the model on top of it.
The company is also recruiting from the music industry. Ethan Kaplan, former Chief Digital Officer at Universal Audio and Fender, is joining to lead Stability’s professional music product suite, details of which have not yet been disclosed. ElevenLabs, its closest licensed competitor, similarly hired Derek Cournoyer from Kobalt as Strategy Lead for Music Business Affairs earlier this year, and Suno hired former Merlin CEO Jeremy Sirota as chief commercial officer. The pattern suggests music industry relationships are now a hiring priority across the entire AI audio sector.
Training on licensed data resolves legal exposure; it does not guarantee model quality or commercial relevance. The open-weight Stable Audio 3.0 models will be widely tested by developers and producers over the coming weeks, and real-world output quality on complex compositions will determine whether the licensing advantage translates into actual market share.
Google and ElevenLabs are also releasing models and tooling in the AI music space, but neither has Stability AI’s combination of open-weight availability, label partnerships at both UMG and WMG, and a dedicated professional music product roadmap launching simultaneously.
As AI music generation moves from novelty to production tool, the companies that control licensed training data and professional workflow integrations will define the next competitive layer, a race Relve, an AI trends intelligence platform, is tracking across the creator economy stack.
