Suno's New Music AI Is Trained on Songs It Paid For

Suno released its sixth generation of music models on September 9, and the most important thing about them is not the sound. It is the training data. Suno says v6 was built with licensed music from Warner Music Group, BMG and the distributor Believe, with revenue shared back to the rightsholders.

That is a sharp turn for a company still defending itself in court. The major labels have spent two years suing generative music tools over training on copyrighted recordings without permission, and Suno was among the targets. Building a flagship model on licensed material, rather than scraped audio, is an attempt to move from defendant to business partner.

Three models, one of them free

The v6 lineup has three versions, according to Axios. A base v6 model aims for controlled, reliable output. A v6-wild variant is tuned for more experimental, unpredictable results. And v6-mini, a faster model, is free to every user. Suno says it trained v6 from the ground up on a mix of licensed Warner recordings and audio generated by its own users, and that it will retire its older models once v6 has rolled out.

Paying subscribers on Suno's $8 Pro and $24 Premier plans get the flagship model and the experimental one. The financial terms of the label deals were not disclosed, though Digital Music News reported that the agreements include revenue sharing with rightsholders as the models are used.

A licence on one front, a lawsuit on another

The timing is pointed. Even as Warner licenses its catalogue to Suno, music publishers are pressing copyright claims elsewhere in the industry, including a suit by Sony and Warner Chappell against Anthropic over song lyrics used to train Claude. The split shows an industry hedging: license where a deal can be struck and the economics work, litigate where they cannot.

For Suno, the gamble is that paying for training data buys something scraping never could, which is a product the labels will promote rather than sue. Whether licensed data narrows the creative range of what the model can produce, and whether artists see a meaningful cut of the revenue, are the questions that will decide if this becomes the template for AI music or a one-off truce.