Stability AI unveils audio model capable of generating songs up to six minutes long

Stability AI, the artificial intelligence company best known for developing Stable Diffusion, has introduced a new family of audio-generation models under the Stability Audio 3.0 brand. According to the company, the most advanced model in the lineup can produce professional-quality musical compositions exceeding 6 minutes in length, marking a significant advancement in AI-generated audio.

The newly announced Stability Audio 3.0 family comprises four models designed to serve different use cases and performance requirements. The lineup includes Small SFX with 459 million parameters, Small with 459 million parameters, Medium with 1.4 billion parameters, and Large with 2.7 billion parameters.

The two smaller models are designed primarily for on-device sound and music generation. Stability AI says these are optimised for efficiency and can create audio clips up to 2 minutes long. Their lightweight architecture makes them suitable for applications where local processing and lower computing requirements are important.

The Medium and Large models are significantly more powerful and are designed to generate complete musical pieces. According to the company, both versions can produce compositions lasting up to six minutes and twenty seconds while maintaining coherent musical structure, rhythm, and melodic consistency throughout the track.

This represents a substantial increase over the capabilities of Stability Audio 2.0, introduced in 2024. The new generation more than doubles the maximum audio length users could previously generate, allowing creators to produce longer, more complex musical works with fewer interruptions or structural inconsistencies.

As part of the launch, Stability AI is making several of the models openly accessible. The Small SFX, Small, and Medium variants are being released with open weights, allowing developers, researchers, and creators to customise and modify the models for their own projects.

The move continues Stability AI’s broader commitment to open-source AI development. In 2024, the company launched Stable Audio Open, which enabled users to generate music clips up to 47 seconds long. Compared with that release, the new Stability Audio 3.0 family represents a major leap forward in both duration and overall capability.

The Large model, however, will not be distributed under the same open-access framework. Instead, it will be available exclusively through the company’s application programming interface (API) and paid self-hosting organisations seeking access to the most advanced model will need to use Stability AI’s commercial offerings rather than downloading the model directly.

The company also noted that businesses generating more than $1 million in annual revenue will be required to obtain an enterprise license to use the technology commercially.

The launch comes at a time when competition in AI-powered music generation is rapidly intensifying. Several major technology companies and startups are investing heavily in tools that use generative AI to create songs, sound effects, and complete audio productions.

Companies such as Google and ElevenLabs have introduced their own music-generation technologies and creative audio tools to capture a share of the emerging market. At the same time, the industry continues to face important legal and licensing challenges. Ongoing court disputes involving Suno and Udio have highlighted concerns about the use of copyrighted music for AI training. These legal battles have underscored the importance of data licensing agreements and partnerships with music rights holders for companies operating in the generative music space.

As AI-generated music becomes increasingly sophisticated, relationships with record labels and content owners are expected to play a critical role in determining which companies can sustainably scale their products while avoiding legal issues. Recognising the importance of content, Stability AI has already taken steps to strengthen its position in the music industry. Last year, the company announced partnerships with Warner Music Group and Universal Music Group to develop AI models and music-creation technologies.

According to Stability AI, the latest Stability Audio 3.0 models have been trained exclusively on fully licensed datasets. The company says this approach is intended to ensure responsible model development while addressing concerns surrounding intellectual property rights and copyrighted music.

Beyond releasing new models, Stability AI is also developing a broader suite of products specifically for professional musicians and music creators. While the company has confirmed that development is underway, it has not yet disclosed detailed information about the upcoming tools or their specific capabilities.

To support these efforts, Stability AI has brought in experienced leadership from the music technology industry. Ethan Kaplan, who previously served as chief digital officer at Universal Audio and Fender, is joining the company to oversee and lead its professional music initiatives.

The appointment reflects a growing trend among AI companies seeking to strengthen their music-industry expertise by recruiting experienced executives from established entertainment organisations.

Earlier this year, Suno organisation appointed CEO Jeremy Sirota as chief commercial officer. Similarly, ElevenLabs recruited Derek Cournoyer from independent music publisher Kobalt Music Group to help guide strategy for its music-related business operations.

With the launch of Stability Audio 3.0, Stability AI is further expanding its presence in the rapidly evolving generative audio market. By introducing longer-form music generation, offering open-weight models for developers, leveraging licensed training data, and investing in professional music products, the company is strengthening its position in a sector where technological innovation, commercial partnerships, and copyright compliance are becoming increasingly important.