TL;DR
On 9 September 2026 Suno launched v6, built in partnership with Warner Music Group, BMG and Believe, and said in the same post: "As v6 rolls out, we will retire our previous models and move Suno entirely onto the v6 generation." There is no rollback. If your prompts stopped producing the vocal sound you had, that is the model swap, not you. This was not a surprise: the Warner settlement in November 2025 already committed Suno to licensed models and to deprecating the old ones. The takeaway for producers is not that AI got worse, it is that the voice you were building tracks around was never yours to keep.
If you woke up to a Suno that does not sound like the Suno you paid for, you are not imagining it. The model you were using is being switched off.
This is worth understanding properly, because it is not a bug, a bad rollout, or something that gets patched next week. It is the deal Suno signed to stay in business, and it changes the calculation for anyone who was using AI vocals as a serious part of their workflow.
What Actually Happened on 9 September 2026
Suno released a new model family and announced it with the headline "Introducing v6: A New Generation of Music Models, in Partnership with the Music Industry". That subtitle is the whole story.
Three models: v6 (the polished default), v6-wild (more unpredictable and textured), and v6-mini (faster, and the only one free users get).
Named label partners: Warner Music Group, BMG and Believe.
Artist participation: "opt-in experiences built around individual artists, where artists can choose to participate and get paid when they do."
The line most people missed: "As v6 rolls out, we will retire our previous models and move Suno entirely onto the v6 generation."
Read that last one again. Not "we recommend v6." Not "v6 is the new default." Retire, and move entirely.
Why Suno Did This
Because they had already promised to. In November 2025, Warner Music Group settled its copyright lawsuit against Suno in exchange for a licensing partnership. Music Business Worldwide reported the terms at the time: Suno would launch "new, more advanced and licensed models," and "the current models will be deprecated."
That is exactly what just happened, ten months later. The reason the old models had to go is uncomfortable but simple: they were trained on material Suno did not have permission to use, so they could not be part of a licensed product.
Eight days before v6 launched, that stopped being an allegation. In its 1 September 2026 answer to Universal and Sony's amended complaint, Suno admitted that audio data was obtained from YouTube for use as training data using yt-dlp, while arguing the labels lack standing to bring the stream-ripping claim. We covered the mechanics of that practice in training AI on YouTube rips.
Where the litigation stands
| Date |
What happened |
| Jun 2024 |
Warner, Universal and Sony sue Suno over training on unlicensed recordings. |
| Nov 2025 |
Warner settles. Suno commits to licensed models, artist opt-in, and deprecating the current models. |
| Aug 2026 |
Judge keeps the Universal and Sony case at 560 recordings, refusing an expansion to 61,026, but allows a DMCA stream-ripping claim. |
| 1 Sep 2026 |
Suno admits it pulled YouTube audio with yt-dlp for training data. |
| 9 Sep 2026 |
v6 launches with Warner, BMG and Believe. Every previous model is scheduled for retirement. |
Universal and Sony have not settled. That case is still live in Boston. For the wider picture, see what the Suno ruling means for producers and Suno's loss to GEMA in Germany.
Is v6 Actually Worse?
Honest answer: it is days old, and the verdict is still forming. Anyone publishing a confident measurement of v6 versus v5.5 right now is guessing.
What is not a guess is the mechanism, and it explains why so many producers are hearing a downgrade rather than an upgrade.
A licensed model is a constrained model. Anything trained on scraped catalogue could imitate an enormous range of singers. A model that can only draw on cleared material has a smaller pool to draw from, by design.
Artist voices are now opt-in. If the singer whose texture your prompt was leaning on has not opted in, that texture is not in the model any more. No prompt gets it back.
Your prompt library is calibrated to a model that no longer exists. Every prompt you refined over months was tuned against specific model behaviour. That behaviour is being switched off.
So the useful question is not "is v6 worse." It is: can you build on something that can be replaced overnight by a settlement you were never party to? Because that just happened once, and Universal and Sony are still litigating.
What Producers Lose in Practice
-
No rollback. The models are being retired, not archived. You cannot pin a version.
-
Free tier cannot download. Under the Warner terms, free users play and share. They do not download.
-
Monthly download caps on paid tiers, with extra downloads sold separately.
-
Voices come and go with opt-in status. The pool of available vocal character is now a business decision, updated without notice.
-
Your released catalogue is unrepeatable. Want a follow-up single that sounds like the one that did well? The model that made it is gone.
That last point is the one that bites hardest. Consistency is what turns tracks into an artist project, and it is the exact thing a retired model takes away from you.
The Problem That Was Always There
Strip out the news and one fact has been true the whole time: with an AI vocal, you never owned the voice. You owned an output, subject to terms of service that can change, from a company whose training data is being argued over in federal court.
Compare that with a licensed acapella from a real vocalist.
|
AI generated vocal |
Licensed acapella |
| Still available next year |
Only if the model survives |
The file is on your drive |
| Same voice on the next track |
No guarantee |
Yes, same vocalist |
| Written licence you hold |
Terms of service, revisable |
Yes |
| Exclusivity possible |
No |
Yes, on exclusive terms |
| A human to credit and promote |
No |
Yes |
If you are weighing this up properly, exclusive vs non-exclusive vocals explains which licence actually fits what you are building.
What to Do Now
You do not have to pick a side in the AI argument to make a practical decision this week.
1. Export everything you still can. Anything you generated and want to keep, get it off the platform onto your own drive while your current downloads last.
2. Do not restart a project on a model in transition. Starting a new EP on v6 today means betting on it still sounding the same when you finish.
3. Use AI where it is genuinely good: sketching a topline, testing a melody, deciding whether an idea is worth finishing. Then track the real thing.
4. Get one real vocal in your next release. A licensed acapella, or a vocalist you hire directly. Compare the two in the same session and judge it with your own ears.
Two routes to a real vocal, depending on how much control you want. Licensing a ready acapella is instant and cheap, covered in how to find acapellas. Commissioning something written for your track is covered in how to hire a vocalist, with real costs.
If you still want to run both, how to humanize AI vocals covers getting more out of what you generate, and Suno vs Udio vs real vocalists puts the three side by side.
Frequently Asked Questions
Can I still use Suno v5 or v4.5?
Not for much longer. In the v6 announcement on 9 September 2026 Suno stated it will retire its previous models and move the platform entirely onto the v6 generation. There is no published option to pin an older version, and model deprecation was part of the Warner Music settlement agreed in November 2025.
Why does Suno v6 sound different from the model I was using?
v6 is a new model family built in partnership with Warner Music Group, BMG and Believe, with artist participation on an opt-in basis. A model trained on cleared material draws on a narrower pool of vocal character than one trained on scraped catalogue, and prompts calibrated against the old model behaviour will not reproduce the old results.
Is Suno legal to use now?
Suno's position improved with the Warner settlement, but it is not settled. Universal Music and Sony Music are still litigating in Boston, and in September 2026 Suno admitted it obtained YouTube audio using yt-dlp for training data while disputing the labels' standing on that claim. Read the terms of your own tier before releasing commercially.
Can I download my old Suno tracks before the models are retired?
Download what you want to keep now. Under the Warner deal terms, free accounts can play and share but not download, and paid accounts have monthly download caps with additional downloads sold separately.
What is the alternative to AI vocals for producers?
A licensed acapella from a real vocalist. You get a dry studio file on your own drive, a written licence, the option of exclusivity, the same singer available for your next track, and a human to credit and promote alongside the release.
Does a licensed acapella cost more than a Suno subscription?
A single non-exclusive acapella licence usually costs less than one month of a paid AI subscription, and you own the file permanently rather than renting access to a model that can be retired.
The Short Version
Suno is doing the responsible thing by moving to licensed models. It is also proving the point that has been sitting under this technology the whole time: a voice you rent can be taken back, and a voice you licence from the person who owns it cannot.
Your prompts are gone. Your vocalist is not.
A voice that cannot be deprecated
Real vocalists, dry studio acapellas, a written licence, and the same singer available for the next track. Preview before you buy.
Browse Real Acapellas
