Suno, which generates music with AI, has released its new model, v6. What stands out in the company's telling is not what the model can do but what it was trained on.
v6 is the first Suno model built with the support of the record industry. After two years of copyright litigation, that is the most visible shift in the company's stance.
The data was rebuilt from scratch
Suno's Jack Brody told The Verge that v6 was "trained from the ground up, with a new set of data that does not include the same data that our previous models were trained on". The set includes content licensed from Warner Music Group, BMG and Believe, along with what the company calls "user data".
That phrasing leaves a significant gap. The new model's data may be clean, but that does not settle how the earlier models were trained. And the earlier data is precisely what the lawsuits against Suno are about.
The music side of the deal is not one-directional either. In licences of this kind, labels usually ask for both a revenue share and limits on how an artist's name may be used. Suno did not share the detail of those limits; it only said the data is licensed.
Three separate models
v6 is not a single release. The company offers three variants, split across different jobs.
| Version | Focus | Access |
|---|---|---|
| v6 | Standard production quality | Paid plans |
| v6-wild | More experimental, unpredictable output | Paid plans |
| v6-mini | Speed and low cost | Free to all users |
Making the free tier the mini model is familiar product tiering. The real difference sits on the licensing side: for commercial use of an output, which version produced it is now a meaningful question.
Shipping three versions at once also shows the user base has split in two. On one side is a broad free audience running quick experiments; on the other, a smaller group using the output at work and wanting predictability.
What licensed data changes
A model trained on licensed data carries less risk for anyone using its output commercially. Even so, licensing alone does not settle who owns the track that comes out.
- The agreement covers training data only, not the usage rights to every track generated.
- The three companies have large catalogues but do not represent the whole industry.
- How independent artists' recordings sit in the set has not been explained.
Another open point is the term of the licence. Training-data licences are usually renewed periodically, and when they are not, the model's status becomes unclear. Since the detail of such contracts is not made public, the only measurable thing for a user is the company's own statement.
Seen from a working studio
In advertising, games and video work, the copyright history of the music used is asked about more and more often. Content-matching systems on publishing platforms can raise claims against tracks of unclear origin.
So the practical advice is simple: keep a record of which model produced the track, which plan it was produced under, and what that plan said about commercial use. Suno's step with v6 makes keeping that record easier, not unnecessary.