Stability AI's Stable Audio 3.0 extends music generation to six-minute compositions
Stability AI releases four new audio models capable of generating full-length songs, with open-weights tiers and licensing deals backing the release.
Last verified:
Stability AI introduced Stable Audio 3.0 on May 20, a four-model audio generation family with dramatically expanded capabilities. According to TechCrunch AI, the largest variant can synthesize complete musical works lasting 6 minutes 20 seconds while preserving melodic structure and harmonic consistency—more than double the generation length of its 2024 iteration.
The release strategy balances open accessibility with commercial control. Stability AI is releasing three tiers as open-weights models: the 459M-parameter SFX variant, the 459M-parameter general-purpose model, and the 1.4B-parameter mid-tier configuration. The 2.7B-parameter flagship remains proprietary, accessible exclusively through paid API endpoints or self-hosting arrangements. Companies with revenues exceeding $1M must negotiate enterprise licensing terms.
Scale and capability progression
The parameter progression reflects deliberate trade-offs between resource consumption and output fidelity. The smaller models target on-device deployment, constrained to 120-second generations. The medium-tier and large configurations unlock the full-composition capability, enabling producers to generate works of concert-like duration. This represents a substantial leap from Stable Audio Open (released in 2024), which peaked at 47-second generation windows.
Licensing and music-industry partnerships
Stability AI’s foundation for Stable Audio 3.0 rests on fully licensed training data, supporting the company’s positioning in an increasingly contentious IP environment. The company executed deals with Warner Music Group and Universal Music Group in 2025 to co-develop creation tools and models. This contractual footing contrasts sharply with ongoing litigation faced by competitors Suno and Udio, whose training practices have drawn copyright challenges.
Talent recruitment and competitive positioning
Stability AI appointed Ethan Kaplan—former chief digital officer at Universal Audio and Fender—to oversee professional-musician offerings. This hiring reflects a broader industry shift toward recruiting music-business veterans. TechCrunch AI notes that Suno recruited Jeremy Sirota (ex-Merlin CEO) and ElevenLabs brought on Derek Cournoyer (from indie publisher Kobalt) in parallel talent acquisitions.
Why This Matters
The release signals that music-generation vendors are moving from novelty demonstrations toward production-grade workflows. Teams evaluating music-AI tools must now weigh open-weights flexibility (available immediately for the small and medium tiers) against enterprise feature completeness (the large model’s exclusivity). For independent musicians and smaller studios, the open-weights deployment removes infrastructure barriers; for labels and publishers, the licensing agreements with major rights holders may become a deciding factor in tool selection as litigation outcomes reshape the industry’s data-sourcing standards.
Frequently Asked Questions
What are the parameter sizes of the new Stable Audio 3.0 models?
The lineup includes two 459M-parameter variants (small SFX and small), a 1.4B-parameter medium model, and a 2.7B-parameter large model. Smaller versions generate up to two minutes of audio; larger models produce compositions up to 6 minutes 20 seconds.
Which Stable Audio 3.0 models are publicly available?
According to TechCrunch AI, Stability AI is releasing the small SFX, small, and medium models with open weights. The large model is restricted to paid API and self-hosting tiers, with enterprise licensing required for companies exceeding $1M in annual revenue.
How does this compare to Stable Audio 2.0?
Stable Audio 3.0 can generate music more than twice as long as Stable Audio 2.0 (released in 2024), extending from previous maximum lengths to full-composition durations while maintaining structural coherence and melodic consistency.
What licensing approach does Stability AI use?
Stability AI trained Stable Audio 3.0 on fully licensed data and signed agreements with Warner Music Group and Universal Music Group in 2025 to develop music models and creation tools.