Suno replaces its models with one trained on licensed music amid lawsuits

Suno, one of the best-known AI music generation startups, announced on Wednesday a new family of models, Suno v6, trained on licensed data from labels and distributors such as Warner Music Group, BMG, and Believe. According to the company, v6 was not trained on the data used in previous versions, the same data that triggered a wave of copyright infringement lawsuits.

The move isn't purely technical: it's a response to litigation. And that's where the news matters to those who build software, not just those who make music.

What came before

Suno had been sued by record labels that accuse the startup of training its models on copyrighted material. The company settled with Warner Music Group last year and settled with BMG last month, according to TechCrunch. The announcement of the new model came, in fact, one day after the company admitted to having trained its models using YouTube videos.

In other words: Suno v6 is the product of a strategy shift. Instead of defending its previous training approach, Suno began licensing catalog and rebuilding its dataset. It's a trade of legal risk for licensing cost, something any team that trains or fine-tunes models on third-party data should watch closely.

What Suno launched

There are three versions, with distinct positioning:

| Model | For whom | Use | |---|---|---| | Suno v6 | paying users | base model, described as reliable and controllable for predictable outputs | | Suno v6 wild | paying users | experimental, aimed at ideation and unexpected results | | Suno v6 mini | all users | faster version, available to everyone |

The company plans to retire the older models. Among the new features: editing part of a song via prompt or a word from the lyrics, using text, image, or video as reference to create tracks, and separating an instrument from a sample to build a new beat.

Suno also said it plans to add song remixing, but conditional on an opt-in program set up with the labels, in which artists authorize the use of their songs in AI-generated features. According to Jack Brody, the company's chief product officer, the logic is to open up new revenue streams.

I think the music ecosystem and our partners are always looking for ways to create more revenue opportunities for rights holders and artists. So a big part of this launch is creating additional revenue streams there.

>

-- Jack Brody, chief product officer at Suno

Last month also brought other governance measures: Suno said it will add watermarks to songs generated on the platform and introduced download limits by account tier.

The lawsuits aren't over

Despite v6, Suno still faces lawsuits from Sony and Universal Music Group, as well as suits from artists like Jason Isbell and from users who allege the company ignored safety in pursuit of profit. Even amid the controversy, the startup has already raised more than $819 million, according to PitchBook data cited by TechCrunch.

It's important to separate two things: Suno v6 may have clean training, but that doesn't erase the legal exposure of previous models or the outputs they've already generated. For those who built products on top of the old API, this distinction isn't academic.

What changes for those building in Brazil

Many Brazilian startups and agencies use audio generation tools as a layer within their own product: soundtracks for videos, jingles, app sound design, content generation at scale. The Suno case reinforces a point that often slips through the roadmap: the provenance of the training data behind the model you embed is a risk that transfers to your product.

Some practical takeaways from the episode, from my perspective following the ecosystem:

• A model with "clean" training becomes a contractual differentiator. Suno is turning "trained on licensed data" into a sales pitch. Expect larger clients in Brazil to start demanding this guarantee in writing from generative AI vendors, especially for audio and image.
• Watermarks and download limits change integration. If you consume Suno's output via API, audio watermarking and new tier-based limits may affect production pipelines and volume. It's worth checking the terms before scaling.
• Artist opt-in creates external dependency. Features like remixing only work for catalog that has opted to participate. A product that promises "remix any song" may not have guaranteed coverage.
• Retiring old models breaks reproducibility. Suno will retire previous versions. If your product depends on a specific model's behavior, a forced migration from v5 to v6 can change outputs without useful warning, the classic problem of depending on a third-party-hosted model.

What remains open

Suno hasn't detailed the volume or exact scope of the licensed catalog feeding v6, nor how liability plays out for content already generated by the old models. It's also unclear how the opt-in program will cover regional catalogs, which matters for the Brazilian music market. And the Sony and Universal lawsuits are still ongoing, meaning the legal foundation of the music AI sector still hasn't settled.

For the developer building on top of this, the lesson is less about Suno and more about posture: treat the provenance of a third-party model's training data as part of your technical due diligence, at the same level as latency, cost per call, and SLA. Today's dirty training data is tomorrow's lawsuit against your client.