ElevenLabs has released Music v2.5 for ElevenMusic and made it available both in the app and through its API, while the previous v2 model stays accessible. The company says generated songs now sound fuller and more natural. In a blind test of 47,885 comparison pairs, listeners preferred v2.5 in most cases, with the strongest results in R&B, Soul, Hip-Hop, Rock and orchestral music. For businesses that put generated audio into products, the release matters less as a quality claim and more as a change in access terms: the model is now reachable through the same API that already serves other ElevenLabs products.
What the new tiers allow
The free tier includes five lossless downloads per day, and the Pro tier includes 400 per month. Users keep the rights to the tracks they generate. Commercial use is allowed, but it depends on the industry and the purpose, and on the free tier attribution is required. Two restrictions are written into the product itself: downloads of tracks based on other artists' songs are blocked, and imitating existing musicians is not allowed. Those limits apply regardless of the plan, so a paid subscription does not remove them.
Technically, the release keeps two model versions running side by side. Music v2.5 is offered through the app and the API, while v2 remains available, which means existing integrations do not have to be rebuilt on release day. That is a deliberate choice for teams that have already wired a specific model version into a production pipeline: they can test v2.5 on a separate branch, compare output against their current results, and switch when the quality difference justifies the work. The blind test of 47,885 pairs is the company's own evidence for that comparison, and it covers genre preferences rather than a general quality score.
The licensing background is where the release gets more complicated. ElevenLabs recently signed a licensing deal with Universal Music Group, but the company says the agreement applies only to future, separate products and not to Music v2.5. ElevenLabs states that it trained its existing Music models on licensed stems and music, without detailing the data. That position separates it from competitor Suno, which was sued for training on copyrighted content without rights holders' consent. Suno also released new models this week, so the two vendors are moving in parallel, with different answers to the same question about training data.
What this means for business
For companies that use generated music in advertising, games, video or internal content, the practical change is the download quota and the API route. Five lossless files a day on the free tier is enough for testing and small campaigns, but not for a content operation with a regular release schedule; 400 a month on Pro sets a clearer ceiling for planning. Attribution on the free tier adds a step to publishing workflows, and the industry-and-purpose condition means legal review is still needed before commercial placement. A small team can validate the model on the free tier; a larger one will have to budget for Pro and for the review time.
What the release does not settle is the rights picture around training data. ElevenLabs says its models were trained on licensed stems and music but does not detail the dataset, and the Universal Music Group deal does not cover Music v2.5. Before choosing a vendor, a buyer should ask which model version is covered by which agreement, what attribution is required in each tier, and how the ban on imitating existing musicians is enforced in practice. The blocked downloads of tracks based on other artists' songs are a product-level control, not a legal guarantee, so the responsibility for the final use stays with the company that publishes the track.
The marker to watch is whether ElevenLabs extends the Universal Music Group agreement to the Music line itself, or keeps the licensing deal tied to separate future products. If the coverage expands, the training-data question becomes less of a procurement risk for companies in music-adjacent industries; if it stays narrow, buyers will keep treating each model version as a separate legal item and will ask for documentation before putting generated tracks into commercial distribution.
