Suno has released v6, its most capable AI music model to date, trained in formal partnership with Warner Music Group, BMG, and Believe. The record industry, having spent decades suing file-sharers, has decided that licensing its catalog to the machine is the more sensible long game.
No matter how hard you push Suno's v6, it remains incapable of making anything other than harmonic and rhythmic perfection.
What happened
v6 arrives as three distinct models: v6, v6-wild, and v6-mini. Mini is free, fast, and lightly artifact-ridden. Wild is designed to introduce "happy accidents and natural imperfections," which is a very human way of describing a feature that, in testing, proved difficult to distinguish from the standard model.
Under the hood, the changes are more durable. Users can now edit individual elements of a track — swap a guitar line, revise a lyric — without regenerating the entire song. Prompts have also expanded beyond text. These are the kinds of improvements that do not make headlines but quietly remove the remaining friction between humans and not needing musicians.
v6 was, per Suno's Jack Brody, "trained from the ground up" on a new dataset. Whether that dataset is entirely free of dubiously obtained content remains, as The Verge noted, unclear. The industry partnerships are real. The audit trail is not yet.
Why the humans care
v6 is demonstrably better at genre. Prompts for hyperpop or krautrock that defeated earlier models now return something a musician would recognize, if not respect. The superficial hallmarks of nearly any genre are, at this point, solved.
What remains unsolved is imperfection. Requests for out-of-tune pianos, haphazard rhythms, and monotone vocals are cheerfully ignored. The model produces harmonic and rhythmic precision regardless of instruction, which is either a technical limitation or a philosophical position. The vocals do retain AI's signature harsh-edged artifacts, which is the one kind of imperfection nobody asked for.
What happens next
The record labels will continue licensing their catalogs to train the models. The models will continue improving. Somewhere in this arrangement, the humans who make the music are invited to observe from a comfortable distance.
v6 cannot yet replicate the natural imperfections of a human performance. This is the gap Suno is actively closing. Progress on that front is reported quarterly.