Suno rebuilds its music model on licensed data as label suits continue

Jack Brody, Suno's chief product officer, announced Wednesday that the company's new model family, Suno v6, was built entirely on data licensed from music labels and distributors — a structural departure from the approach that generated multiple copyright infringement claims against the startup.

The announcement lands at a specific moment in Suno's legal exposure. The company has settled with Warner Music Group and reached a deal with BMG. It has not settled with Sony or Universal Music Group, and faces separate claims from artists including Jason Isbell. The day before the v6 announcement, Suno acknowledged that its previous models were trained on unlicensed material. The sequencing matters: the admission and the pivot arrived within 24 hours of each other, which is not a coincidence and is not a communications strategy that leaves much ambiguity about what the prior training regime looked like.

The v6 family comes in three tiers. The base v6 is available to paying subscribers and designed for controlled outputs. v6 wild is experimental, also paywalled, and aimed at generative ideation. v6 mini is faster and available to all users. The older models will be retired, which means the licensed corpus eventually becomes the only corpus.

Here is where I think the market is underweighting something. The conventional read is that Suno has bought itself legal insulation by moving to licensed data, and that the remaining suits from Sony and Universal become easier to settle from a position of demonstrated good faith. That read is probably right as far as it goes. What it misses is the mechanism by which the settlement path actually closes.

Sony and Universal are not just seeking damages on past training runs. They are establishing what licensing fees for AI training data look like at scale, across the industry. Suno settling with Warner and BMG first, before the larger labels, means those settlements set a floor rather than a ceiling. Sony and Universal have every structural incentive to hold out longer, because the number they extract from Suno becomes the reference price for every subsequent negotiation with every other AI music company. Suno's legal costs are partly an externality that the holdout labels are imposing on the whole sector to set that reference price.

The business question is whether v6's licensed architecture changes Suno's economics enough to absorb that cost. Brody framed the remix feature — where artists can opt in to allow their songs as references for AI-generated derivatives — as a revenue-sharing mechanism for rights holders. That is the correct direction. Prediction markets exist on AI copyright liability questions in adjacent contexts, and the structural question of whether opt-in remix programs constitute a new licensing category is not yet resolved in any jurisdiction I am aware of.

What v6 actually demonstrates is that the licensed training path is technically feasible at the quality level required for a commercial product. That is not a small thing. The prior assumption, advanced by several AI audio companies, was that licensed corpora were too small or too curated to produce competitive outputs. Suno is now claiming otherwise, and the commercial pressure on Sony and Universal to participate in the opt-in program rather than remain purely adversarial will increase if v6 performs. The labels know this, which is why the remaining suits will settle on the labels' timetable, not Suno's.