Stability AI Introduces New Audio Models Capable of Generating 6-Minute Music Tracks

Stability AI launches new audio models that can generate 6-minute music tracks

May 21, 2026 By Mandy Dalugdug

Stability AI has launched Stable Audio 3.0 – a new family of four AI music models that the company says are trained entirely on licensed data and can generate tracks of more than six minutes in length.

Three of the four models are open-weight, meaning they are free to download and build upon.

The launch, announced on Wednesday (May 20), represents a significant step up from Stable Audio 2.0 , which launched in April 2024 with a maximum generation length of three minutes.

Stability AI said in its announcement: “Today we’re releasing Stable Audio 3.0, a model family trained on fully licensed data, designed to be the foundation for what the audio community builds next.”

“Three of the models are open weights, free to download and build on.”

The company added: “Music has always evolved through the collective creativity of its community.

> “Generative audio will be no different. We want to foster the same kind of community-driven innovation in audio that we sparked in image generation with the launch of Stable Diffusion.” Stability AI

“Remix culture, interpolations, and mashups are how artists build on each other’s work and push the art form forward.

“Generative audio will be no different. We want to foster the same kind of community-driven innovation in audio that we sparked in image generation with the launch of Stable Diffusion.”

The four models released under the Stable Audio 3.0 banner are: Small SFX , designed for sound effects generation on mobile phones and consumer-grade laptops; Small , for full music composition on-device; Medium , offering longer track lengths of up to 6 minutes and 20 seconds ; and Large , which Stability AI says is its most advanced model, built for music platforms and creative applications requiring low-latency generation at high volume.

The Small SFX and Small models each have 459 million parameters and can generate audio of up to two minutes.

The Medium model has 1.4 billion parameters and the Large model has 2.7 billion .

Small SFX , Small , and Medium are all available as open-weight models on Hugging Face .

The Large model is not open-weight – it is available only via the Stability AI API, through partner fal.ai , or via enterprise licensing for self-hosted deployment.

Stability AI said: “All Stable Audio 3.0 models are trained on fully licensed data.

“Under the Stability AI Community License, you own your outputs and can distribute and commercialize them freely.”

The company added: “To our knowledge, other open music models either restrict commercial use or carry the risks associated with being trained on unlicensed music.”

Organizations with more than $1 million in annual recurring revenue (ARR) require an enterprise license for commercial use, which Stability AI says also includes legal indemnification.

According to a research paper published alongside the launch, the models are trained on a combination of licensed audio from production library AudioSparx – comprising 806,284 audio files – and Creative Commons recordings from Freesound .

The new models run on what the company describes as a novel semantic-acoustic autoencoder architecture, enabling variable-length generation at per-second granularity.

Stability AI says the Small model is, to its knowledge, the only model capable of full music composition on-device – offline and without short sample limits.

The company also announced support for LoRA fine-tuning – an efficient method for customizing models – alongside the open-weight releases, and audio inpainting features that allow users to modify segments of a track or extend compositions.

Stability AI said that it is also developing a new suite of products for professional musicians, though it did not disclose details on features.

According to TechCrunch , Ethan Kaplan – former Chief Digital Officer at Universal Audio and Fender – is joining Stability AI to lead its professional music offering.

The launch arrives amid a wave of AI music companies hiring executives from the traditional music industry.

Earlier this year, Suno hired former Merlin CEO Jeremy Sirota as Chief Commercial Officer .

ElevenLabs also appointed Derek Cournoyer , formerly of indie music publisher Kobalt , as Strategy Lead for Music Business Affairs in January.

Stability AI’s emphasis on licensed training data and its partnerships with major music companies distinguish it from competitors that have faced copyright litigation.

In late 2025, Stability AI struck deals with both Universal Music Group and Warner Music Group to develop AI-powered music creation tools trained on licensed catalogs.

UMG entered into a strategic alliance with Stability AI in October 2025 , with the companies agreeing to co-develop tools powered by responsibly trained generative AI.

Warner Music followed in November 2025, announcing a partnership to develop responsible, artist-friendly AI tools.

Stability AI said in its press release: “While responsibly trained generative AI models are critical, they are not enough on their own.

“Artist-centric AI will only win if the product experience on a licensed platform is better than the experience on an unlicensed platform.”

The company’s emphasis on licensed training data comes as it continues to face copyright litigation in other areas of its business.

A class-action lawsuit filed in January 2023 by illustrators Sarah Andersen , Kelly McKernan , and Karla Ortiz – alleging that Stability AI used their works without permission to train its Stable Diffusion image generator – is still ongoing , with a trial set to begin in September 2026.

In the UK, Getty Images sued Stability AI over the alleged unauthorized scraping of 12 million photos to train Stable Diffusion .

The UK High Court largely rejected Getty ‘s copyright infringement claims in a ruling in November 2025, though it found Stability AI liable for limited trademark infringement.

On the audio side, musician Anders Manga filed a federal copyright infringement lawsuit against Stability AI and its licensing partner AudioSparx in December 2025, alleging that his recordings were used to train Stable Audio without authorization and in violation of a licensing agreement that predated generative AI.

Stable Audio 3.0 ‘s generation ceiling of 6 minutes and 20 seconds places it among a growing number of AI music platforms competing on output length.

Suno , which has emerged as the market leader in consumer AI music generation – claiming $300 million in annual revenue and 2 million paid subscribers – can generate tracks of up to 8 minutes with its v5 update .

ElevenLabs ‘ Eleven Music platform – which launched in August 2025 with licensing deals with Merlin and Kobalt already in place – can generate tracks of up to 5 minutes .

Google ‘s Lyria 3 Pro , launched in March 2026 , generates tracks of up to 3 minutes . Music Business Worldwide

News United States Stability AI Stable Audio

Related Posts

UMG strikes strategic alliance with Stability AI to develop ‘next-generation’ AI music-making tools

_Originally reported by [Music Business Worldwide](https://www.musicbusinessworldwide.com/stability-ai-launches-new-audio-models-that-can-generate-6-minute-music-tracks/)._