ElevenLabs, Stability AI launch new music models

ElevenLabs and Stability AI released new generative-music models this week aimed at creators, developers and brands. ElevenLabs introduced Music v2 with tools for stitching song sections, switching genres inside a single track and re-generating selected segments. Stability AI rolled out Stable Audio 3.0, a family of four models that includes three open-weight variants and supports outputs up to six minutes and twenty seconds.

ElevenLabs presented Music v2 as its second music model. The company described features that maintain melodic and rhythmic continuity across longer pieces, let users assemble intros, verses and choruses section by section, and replace or “inpaint” specific slices of a song without regenerating the whole track. Music v2 can shift between styles within a track, handle fast rap passages and layer non-musical sound effects, and includes expanded multilingual support. The model is available in the ElevenMusic consumer app and ElevenCreative for brands now; early API access is available through the company’s sales team. ElevenLabs also cut API pricing by up to 50% and lowered ElevenCreative self-serve rates by up to 40%. The company reported $500 million in annual recurring revenue in April 2026 and carried a valuation near $11 billion after a February Series D round.

Stable Audio 3.0 ships as four models: a Small SFX model for on-device sound effects, a Small full-music model intended to run locally, a Medium model that produces tracks up to 6:20 and requires stronger hardware, and a Large API-only model for larger organizations. Three models have open weights published on Hugging Face. The two Small models have roughly 459 million parameters and can run without a GPU. The Medium model has about 1.4 billion parameters and can generate its maximum-length output in roughly 1.31 seconds on an H200 GPU. The Large model has about 2.7 billion parameters and is available via Stability’s API to organizations that meet its commercial requirements. Stable Audio 3.0 uses a semantic-acoustic autoencoder that Stability calls SAME, supports LoRA fine-tuning for catalog adaptation, offers single- and multi-segment inpainting and a causal continuation mode to extend tracks, and integrates with ComfyUI for local setups. Stability emphasized per-second generation granularity to deliver exact track lengths.

Both companies said their models were trained on licensed data and announced partnerships with music companies. ElevenLabs reported licensing agreements with Believe, Kobalt and Merlin. Stability announced partnerships with Universal Music Group and Warner Music Group. The announcements come after industry copyright litigation in 2024 that increased attention on whether AI music models use licensed training material.

Suno remains a leading AI music platform by usage and revenue. Reported figures for Suno include a roughly $2.45 billion valuation in late 2025, about 100 million users, an estimated $300 million in annual recurring revenue and about 7 million songs generated per day. Warner Music settled a dispute with Suno in November 2025; other label disputes continue in federal court.

Stable Audio 3.0’s Small and Medium models are available on Hugging Face now, and the Large model is available via Stability’s API. Music v2 is live for ElevenMusic and ElevenCreative users, with commercial tiers available through ElevenCreative and ElevenAPI.