Suno's Licensed Model Meets Its Own Users

We covered the launch of Suno v6 on September 9, 2026, and the lawsuit Sony Music and Universal Music Group filed nine days later. A third story has developed since then, and it comes from Suno's own customers. Within a week of launch, music-tech and industry outlets were reporting a wave of complaints from paying subscribers who said the new, licensed models sound worse than the ones they replaced.

This is the first large-scale public reaction to a major AI music model trained on licensed catalogs, so it is worth looking at closely, with a clear line between what users report, what the company says, and what can be concluded.

What changed on September 9

According to Music Business Worldwide, Suno launched three models: v6 for Pro and Premier subscribers, v6-wild as a less predictable option for the same tiers, and v6-mini for free users. Chief Product Officer Jack Brody said the models were built from music licensed by partners Warner Music Group, BMG and Believe, user preference data (signals about which versions of songs users prefer), and technical improvements. Suno said the training data does not include Universal or Sony content. All earlier Suno models were retired the same day.

CEO Mikey Shulman framed the release as a foundation: "v6 is designed to expand what is creatively possible and lay the foundation to launch new products that open up revenue opportunities across the music ecosystem." Suno also began sharing a portion of revenue with its licensing partners, who decide how to distribute it to rights holders.

What users are saying

On September 17, MusicRadar reported that the reaction on Suno's community forums was overwhelmingly negative. One user described results as "muffled, dull and strangely lifeless," adding that "the vocals are often buried, the high end feels almost completely gone." Others called the output "flatter, soulless and more similar-sounding" and "overproduced, heavily compressed." One wrote: "I'm giving them a month to fix this problem and if not, I'm cancelling."

Digital Music News followed on September 18 with more of the same: "I spent two days trying all the same styles and prompts in V6 and found it produced muffled, generic, soulless garbage," one subscriber wrote; another said, "I haven't been able to generate a single track I like." Gearnews reported further complaints that prompts were being ignored or lyrics altered, and that the v6-wild and v6-mini variants fared no better with some users.

The retirement of older models sharpened the anger. Users who preferred earlier versions had no way to return to them, and many asked Suno to restore access.

Why it may be happening: claims and limits

The most common explanation among users is the training data. MusicRadar noted that Suno previously described its training material as "all music files of reasonable quality accessible on the open internet." A model trained on catalogs from three licensing partners plus user signals has, users argue, heard less variety. Digital Music News presented this as the likely cause.

That explanation is plausible, but it is not established. Audio quality depends on many things besides the training set, including model architecture, how the system was tuned on preference data, and post-processing. Suno has not published technical details that would let outsiders separate those factors. We could not find a public statement from Suno directly addressing the quality complaints as of October 4, 2026.

The question of what v6 learned from is also in court. Sony and Universal's September 18 suit, filed in federal court in Massachusetts, cites 60,202 recordings and alleges that v6 may have inherited knowledge from the earlier, unlicensed models through a process known as distillation. Suno says v6 was trained from scratch and does not use Universal or Sony data. RouteNote summarized both positions.

Our analysis: provenance has a price, and someone pays it

The v6 reaction shows a tension the licensing era has to resolve. Rights holders have argued for two years that training on their work without permission is infringement. Suno's response was to rebuild on licensed data and retire everything else. Its most vocal users are now saying the result is less satisfying. If licensed models consistently sound worse at first, AI companies face pressure from both sides: litigation if they use unlicensed data, churn if they do not.

There are two plausible ways out. Licensed training pools could grow large and varied enough to close the gap, which is what deals like Stability AI's August agreement with all three majors are betting on. Or the market could split, with licensed generators serving professional and commercial uses where provenance matters, and users who care only about sound chasing whatever tool is least restricted. The second outcome would keep the legal fights going for years.

There is also a simpler reading: much of the enthusiasm for AI music came from novelty, and the novelty may be wearing off regardless of the model. The backlash cannot tell us which explanation is right yet.

What it means for songwriters and artists

For artists, the lesson is that training data is now a visible product feature, discussed by consumers as well as lawyers. Human-written songs remain the raw material that every licensed model depends on, and the value of that material is being priced in real time. At Starchild, the catalog is built on songs written by people, every song carries an AI-transparency label, and the style models used to produce alternate versions are licensed, with revenue shared with the artists behind them.

Sources

• MusicRadar via Yahoo Tech: "I'm cancelling my subscription": Suno users slam its V6 update as "muffled, generic, soulless garbage"
• Digital Music News: Suno Subscribers Have a Big Problem with v6
• Music Business Worldwide: Suno v6 AI music models launch in partnership with WMG, BMG and Believe
• RouteNote: Suno v6 faces user backlash and new copyright lawsuit
• Gearnews: After the Big Deal: Suno v6 Disappoints Users and Labels Alike