Suno v6 is its first AI music model trained with licensed record-label data

Suno has released v6, the first version of its AI music generator built with the participation of the record industry. According to The Verge, the model was retrained from the ground up on a dataset that includes content licensed from Warner Music Group, BMG and Believe, alongside what the company describes as "user data."

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." As The Verge points out, that phrasing still leaves open whether the new training set is entirely free of dubiously obtained material.

Three models instead of one

v6 is actually a family of three. v6-mini is the model available free to everyone, tuned for fast, resource-light generation; The Verge found its output simpler and more prone to the telltale artifacts that betray AI origin. v6-wild is positioned as the unpredictable sibling, with Brody saying it is designed for "happy accidents and natural imperfections" — though in the reviewer's limited testing it was hard to hear much difference between wild and the standard model.

Better genre fluency, same polish problem

The clearest improvement across all three variants is genre comprehension. Prompts for styles such as hyperpop or krautrock that older versions missed badly now reliably hit the superficial hallmarks of whatever genre is requested, according to The Verge's Terrence O'Brien.

What the model still cannot do is imperfection on demand. Attempts to produce off-key, out-of-tune or slightly dissonant results failed every time: requests for "monotone vocals" and "no drums" were ignored outright, and repeated demands for a haphazardly out-of-tune piano yielded nothing of the sort. Meanwhile, the vocals seem to carry more of the harsh-edged synthetic artifacts common to AI models than v5 did.

Editing by conversation, prompts beyond text

Two under-the-hood changes could reshape how people actually use the tool. Users can now edit parts of a song through plain-language instructions in a chat, so changing a guitar line or a single lyric no longer requires regenerating the entire track. v6 can also mash together elements from a user's existing Suno library into new creations.

Prompts are no longer limited to text either. The model can generate tracks based on images, video or other audio. The rollout is underway, and Suno says its previous models will eventually be retired.

Why it matters

Training on licensed catalogs changes both the economics and the legal posture of generative music. If a model's foundations come from deals with rightsholders, the exposure that comes from building on unlicensed material shrinks, and record labels move from adversaries toward business partners. Suno's decision to retrain from scratch, rather than patch an existing model, suggests licensed data is becoming a competitive feature in its own right, not just legal hygiene.

The open question is provenance. Brody's wording stops short of confirming the new dataset is completely clean, and The Verge flags that ambiguity directly. For users, the practical takeaway is a tool that understands genre far better but still cannot sound humanly imperfect on command — a reminder that "licensed" and "convincing" are separate problems. And with mini free for all and older models on their way out, v6 is the version most Suno output will come from going forward.