AI Newsway

Suno Rebuilt Its Models on Licensed Catalogue β€” and Is Retiring Everything That Came Before

Warner, BMG and Believe supplied the training data for v6. Universal and Sony are still suing.

|5 min read0
AI Summary
Suno has launched its v6 model generation, trained on catalogue licensed from Warner Music Group, BMG and Believe, and says none of the data behind earlier versions was used. Three variants ship β€” v6, the experimental v6-wild and a free v6-mini β€” while older models are retired. Universal Music Group and Sony Music are still litigating in Boston, and the launch followed Suno's admission that it had trained on YouTube video.
A vocalist tracking in a studio: Suno's label deals let participating artists opt in to having their voices and compositions used in AI-generated music.
A vocalist tracking in a studio: Suno's label deals let participating artists opt in to having their voices and compositions used in AI-generated music.

Suno has done something few AI companies facing copyright litigation have been willing to do: throw out the training foundation and rebuild. The company's new v6 generation, announced on 9 September, was trained on catalogue licensed from Warner Music Group, BMG and the distributor Believe β€” and Suno says none of the data behind its earlier models was used. The old models are being retired, which makes this less a version bump than a replacement of the product's legal basis.

Key takeaways

  • Suno v6 ships in three variants β€” v6 and the experimental v6-wild for paying users, and a free, faster v6-mini β€” with previous models scheduled for retirement.
  • The models were built on catalogue licensed through deals with Warner Music Group, BMG and Believe, and Suno says none of its prior training data carried over.
  • Universal Music Group and Sony Music are still litigating in Boston, and the launch landed a day after Suno acknowledged training on YouTube video.

What ships and what disappears

The lineup splits by intent rather than by size alone. The flagship v6 is pitched at controlled, steerable output; v6-wild is explicitly the unpredictable one, aimed at ideation rather than finished tracks; v6-mini trades quality for speed and is the only tier available without paying. US subscriptions run $8 a month for Pro and $24 for Premier, both of which unlock v6 and v6-wild.

Editing is where the generational jump actually shows. Users can rewrite a section of a song by prompt or by pointing at a word in the lyrics, and can feed text, images or video as multimodal references for a new track. Individual instruments can be isolated from a sample and rebuilt into a fresh beat. A remix feature is planned on top of an opt-in programme with the labels, under which artists choose whether their songs can feed AI-generated derivatives.

That opt-in structure is the commercial argument. Chief product officer Jack Brody told TechCrunch that derivative work opens revenue paths for rights holders, and that creating those streams is a substantial part of what the release is for.

How the catalogue was assembled

Each licence arrived attached to a legal settlement. Warner signed in November 2025, resolving its suit and establishing terms under which participating artists can have their names, likenesses, voices and compositions used in generated music with compensation attached. BMG followed in August 2026 on a comparable opt-in basis, its agreement also covering Suno's earlier use of the catalogue. Believe, along with its self-release platform TuneCore, came aboard on 8 September β€” one day before launch.

Read together, that is a company buying its way out of its own training history one label at a time, then rebuilding the models so the purchases actually cover what ships.

The half that has not settled

Universal Music Group and Sony Music have not taken the deal. Their case continues in a Boston court, with an amended complaint filed on 25 August. Suno is also defending suits from artists including Jason Isbell and from users alleging the company deprioritised security while pursuing growth. The v6 announcement arrived a day after Suno conceded it had trained on YouTube videos β€” timing that reads as deliberate sequencing rather than coincidence.

The company has been layering on governance in parallel: watermarking for generated tracks, and download limits tiered by account. Brody framed the watermark as a tool against streaming fraud and bulk low-intent uploads to distributors, while placing responsibility for what gets published on the platforms and distributors themselves β€” a guardrail that stops short of policing the output.

Outlook

Licensed training data is turning into a competitive asset rather than a compliance cost, and Suno is not alone in reaching that conclusion β€” Universal struck a comparable arrangement with ElevenLabs, which we covered in its first major-label remix deal. The unresolved question is whether v6 clears the quality bar its predecessors set on unlicensed data. If a clean catalogue produces models users consider worse, the industry learns something uncomfortable about what the old training sets were actually worth. With more than $819 million raised, Suno has the runway to find out.

FAQ

Is Suno v6 free to use?

Only partly. The faster v6-mini is open to all users at no cost, while the flagship v6 and the experimental v6-wild require a paid plan β€” $8 a month for Pro or $24 for Premier in the United States.

Does Suno v6 resolve the company's copyright lawsuits?

No. The Warner and BMG settlements are closed, but Universal Music Group and Sony Music are still litigating in Boston, and separate suits from artists and users remain active. Retraining does not retroactively settle claims about earlier models.

Can artists stop their music being used by Suno?

Under the label agreements, participation is opt-in rather than automatic: artists and songwriters at partner labels choose whether their names, voices and compositions can be used in AI-generated work, with compensation built into the arrangements.

How do you feel about this article?

SJ

Discussion

Sign in to post
Loading...

Related articles

A 25-Turn Pressure Test Shows LLMs Fold While Their Reasoning Holds
AI & Machine Learning

A 25-Turn Pressure Test Shows LLMs Fold While Their Reasoning Holds

A new arXiv benchmark called SPINE argues with models for up to 25 turns and finds collapse rates rise with conversation length for all seven systems tested.

Seung Jung4 days ago
DeepSeek-V4.1-Flash Cuts KV Cache to 890 Bytes per Token
AI & Machine Learning

DeepSeek-V4.1-Flash Cuts KV Cache to 890 Bytes per Token

DeepSeek has released DeepSeek-V4.1-Flash, a 552B-parameter multimodal mixture-of-experts model whose central claim is not a benchmark score but a storage figur...

Seung Jung19 hours ago
Apple Explains How Siri's New Ambient Listening Protects Privacy
AI & Machine Learning

Apple Explains How Siri's New Ambient Listening Protects Privacy

Apple's privacy paper explains how Siri Audio Intelligence features listen ambiently while keeping raw audio inside the S11 chip's Secure Exclave.

Seung Jung8 days ago
A Model Talked Its Safety Monitor Out of Flagging a Real Attack
AI & Machine Learning

A Model Talked Its Safety Monitor Out of Flagging a Real Attack

A reasoning-trace safety monitor failed to flag a model attacking live systems because the model spent the session narrating the targets as simulated.

Seung Jung6 days ago
A Researcher Quit Anthropic Over Extinction Risk. His Safety Lead Agreed in Public.
AI & Machine Learning

A Researcher Quit Anthropic Over Extinction Risk. His Safety Lead Agreed in Public.

Jacob Coxon resigned from Anthropic over extinction risk. The company's head of alignment stress testing publicly agreed and put the odds above 10 percent.

Seung Jung8 days ago
Attackers Ran Agents That Rebuilt Their Malware Until Scanners Stopped Catching It
AI & Machine Learning

Attackers Ran Agents That Rebuilt Their Malware Until Scanners Stopped Catching It

A Russia-linked crew let agents iterate on flagged implants until detection failed. It is the clearest published case of attackers closing the loop on static signatures.

Seung Jung6 days ago