Suno releases its first AI music model made with record industry help
Sep 9, 2026, 2:42 PM · The Verge
Suno’s v6 is the first model it says was rebuilt with licensed label data, but the Verge’s hands-on still finds polished sameness where musicians want grit.
Why it matters
Suno just shipped v6 and says it was trained from the ground up on a new dataset that includes licensed material from Warner Music Group, BMG, and Believe, plus user data. That is the first clear public pivot from the lawsuit era toward a label-aligned training story.
If licensed models become the commercial default, the fight over AI music shifts from “was this scraped?” to “who owns the pipeline, who gets paid, and what kinds of songs the model will actually make.” Creators, labels, and platforms all have different answers.
From the desk
We’re glad the industry is trying to put licensed training on the table. Useful creative tools should not depend on a permanent fog around rights. Jack Brody’s claim that v6 does not reuse the same data as prior models is the point of the launch. Still, The Verge notes it remains unclear whether the new corpus is fully free of dubiously obtained material. That ambiguity matters; a partial clean break is not the same as a clean break.
Product-wise, v6 splits into three flavors: standard, wild, and a free mini tier tuned for speed. Genre following improved sharply in Terrence O’Brien’s tests—hyperpop and krautrock prompts that older models botched now land the surface markers. Chat-style local edits, library mashups, and prompts from images, video, or audio are real workflow upgrades. Those are the parts we’d defend as useful AI: faster sketching, less full-track regeneration tax, more ways into a draft.
The downside is aesthetic and cultural. O’Brien could not coax natural imperfections—off-key vocals, haphazard piano, true dissonance—no matter how hard he prompted. Wild mode promised happy accidents; in limited testing it barely differed. Meanwhile harsh AI vocal artifacts seemed more present than in v5. So the model gets better at sounding like a genre brochure and worse at sounding like a band that left mistakes in.
If this becomes the licensed template, expect cleaner legal footing and blander defaults unless competitors compete on controllable messiness. Suno says older models will eventually retire. That raises the stakes: whatever aesthetic v6 encodes may become the paid baseline.
Context
Suno has faced sustained copyright pressure over earlier training practices. Partnering with major rights holders is both a product story and a settlement-shaped survival strategy. v6 begins rolling out now.
Who feels it
- Musicians and producers
- Better genre hits and surgical edits help drafting; expect less luck generating intentionally imperfect or abrasive performances.
- Labels and rights holders
- Licensed training becomes a commercial wedge—and a way to keep AI music inside deals they can meter.
- Competitors
- Pressure rises to match licensed datasets and multimodal prompting, not only raw fidelity.
What to watch
- Whether Suno clarifies how complete the training-data reset really is
- Creator reaction as prior models are retired
- Whether “wild” modes ever deliver controllable imperfection, not just marketing copy