stability ai has officially unveiled the stability audio 3.0 series of audio-generation models, marking a new milestone in ai music generation—its flagship version supports outputting professional-grade compositions exceeding six minutes in length, with complete structure and consistent style.
the newly released model family comprises four distinct architectures: the xfs lightweight and standard small versions with 45.9 million parameters, a medium-sized version with 1.4 billion parameters, and a flagship large version boasting 2.7 billion parameters. among these, the two smaller models are optimized for edge-device deployment, enabling real-time local generation of sound effects and short musical pieces under two minutes. meanwhile, the medium and large models deliver breakthroughs in temporal modeling and structural consistency, capable of generating continuous works up to 6 minutes and 20 seconds long, featuring natural transitions between musical sections, stable tonality, and strong thematic cohesion—more than doubling the maximum duration compared to the previous-generation stability audio 2.0.
the open-source strategy continues to prioritize community-driven innovation: the small sfx model, along with the small and medium-sized versions, have fully released their weights and source code, allowing free download, fine-tuning, and commercial use. however, the most powerful large model remains unavailable for local deployment; it is accessible only via api interfaces and cloud-hosted services, with enterprises generating over one million dollars in annual revenue required to sign a commercial licensing agreement before gaining access.
in terms of data compliance, stability ai has completed key groundwork—establishing strategic partnerships with warner music group and universal music group to ensure that all training data used for stability audio 3.0 originates from legally licensed music libraries, thereby mitigating copyright risks at the source.
at the same time, the company is accelerating its expansion into the professional audio ecosystem, inviting ethan kaplan, former chief digital officer of universal audio and fender, to join and lead the development of a next-generation ai-powered creative tool suite tailored for professional musicians.
