Stability AI releases six-minute audio model

Stability AI said on May 20 that it released Stable Audio 3.0, a new family of music-generation models that can produce tracks up to six minutes long and, in some versions, run on consumer devices. The company said three of the four models are open weights and were trained on fully licensed data, while the largest version is being offered through its API and self-hosted enterprise deployments. The launch lands as AI-generated music is rising quickly on streaming platforms. Deezer said on April 20 that it was receiving nearly 75,000 AI-generated tracks per day, equal to about 44% of daily uploads, up from 60,000 tracks and 39% of daily deliveries in January. ### What exactly did Stability AI release? (stability.ai) Stability AI said Stable Audio 3.0 includes four models: Small SFX, Small, Medium and Large. The company said the Small SFX and Small models are designed for on-device generation, while Medium and Large can create full compositions lasting as long as 6 minutes and 20 seconds. (newsroom-deezer.com) TechCrunch reported that the Small and Small SFX models each have 459 million parameters, the Medium model has 1.4 billion, and the Large model has 2.7 billion. Stability AI said Small SFX, Small and Medium are available with open weights, while Large is limited to paid API and self-hosting access. (stability.ai) ### Why is the six-minute number getting attention? Stability AI said the new family introduces “variable-length generation up to six minutes,” extending its earlier open music systems. TechCrunch reported that the six-minute-plus ceiling is more than double what Stable Audio 2.0 could generate in 2024, and a sharp increase from the 47-second limit in the company’s 2024 open release. (techcrunch.com) The company said the models use a “novel semantic-acoustic autoencoder” to support longer and more flexible audio generation. That matters because consumer-facing rivals including Suno and Udio have pushed the market toward full-song generation rather than short clips, even as those companies have faced copyright scrutiny and licensing pressure. Billboard said Suno and Udio have been central to the industry’s AI music disputes and later settlement talks with major labels. (stability.ai) ### Are AI songs really becoming that common on platforms? Deezer said the answer on its service is yes, at least on the upload side. The company said fully AI-generated tracks now account for roughly 44% of new daily uploads, compared with 39% in January 2026 and 10,000 AI tracks per day when it first launched its detection tool in January 2025. (stability.ai) TechCrunch reported that Deezer said listening remains much lower than upload volume, with AI-generated music accounting for 1% to 3% of total streams on the platform. Deezer also said 85% of those AI-track streams are detected as fraudulent and demonetized. ### What are rights groups and cultural bodies warning about? (newsroom-deezer.com) CISAC, the global umbrella group for authors’ societies, said in a December 2024 study that 24% of music creators’ revenues are at risk by 2028 because of generative AI. The study said cumulative losses for music and audiovisual creators could reach 22 billion euros over five years. (techcrunch.com) UNESCO said in a February 2026 report that global revenue losses are projected at 24% for music creators and 21% for audiovisual creators by 2028. UNESCO said digital revenues now represent 35% of creators’ income, up from 17% in 2018. ### Where can users get the new models now? (cisac.org) Stability AI said Stable Audio 3.0 Small, Small SFX and Medium are available on Hugging Face now. The company said Stable Audio 3.0 Large is available through the Stability AI API and self-hosting for enterprise customers, with enterprise licensing required for organizations with more than $1 million in annual revenue. (stability.ai) (unesco.org)