Suno Ships v6, Its First AI Music Model Built With Record Industry Backing
Suno released v6, its first AI music model developed with record industry support, trained on licensed content from Warner Music Group, BMG and Believe plus user data. The model arrives in three versions, and while genre accuracy improves, it still cannot produce deliberately off-key or dissonant results.
The company said v6 was trained from scratch on a new data set that does not include the data used for earlier models. Suno's Jack Brody told The Verge the training material includes content licensed from Warner Music Group, BMG and Believe, as well as user data. It is not clear from that description whether the v6 training set is entirely free of content obtained by questionable means.
The release splits into three models rather than one. v6-mini is the free version available to everyone, aimed at fast, resource-light generation; its output is simpler and tends to carry more artifacts that mark it as the product of an AI prompt. v6-wild is meant to be less predictable. Brody said it is designed for "happy accidents and natural imperfections," though in limited testing The Verge found it difficult to tell wild apart from standard v6.
All three versions handle genre far better than their predecessors. Prompts carried over from earlier models asking for hyperpop or krautrock tracks, which had previously missed the mark by a wide margin, produced the superficial hallmarks of each style on v6.
What the model still cannot do, according to The Verge's testing, is deliver the imperfections it claims to chase. Attempts to obtain off-key, out-of-tune or slightly dissonant results failed, as they had on v5. Requests for monotone vocals and no drums were ignored, and repeated prompts for an out-of-tune piano played haphazardly with no regard for key or rhythm, fully out of tune with the rest of the song, produced nothing of the sort. The output stayed harmonically and rhythmically precise. The unnatural side of AI music remained: vocals in particular appeared to generate more harsh-edged artifacts than v5 did.
Several changes operate beneath the surface. Users can now edit parts of a track using plain language inside a chat, and altering a guitar line or a single lyric no longer requires regenerating the whole song. v6 can also combine multiple elements from a user's Suno library into new creations. Prompts are no longer limited to text: the model accepts images, video and other audio as inputs.
Suno said v6 begins rolling out today and that its earlier models will be phased out over time. The company did not give a schedule for retiring them.