Suno v5 and v5.5: Every Current Model Is Being Retired

Where the versions actually stand

Search interest still concentrates on Suno v5, which is a year out of date. The current model is v5.5, released on 26 March 2026. Both are scheduled to be switched off.

That last part is the reason this page exists, and it has gone almost entirely unreported. Buried in Suno's 10 August downloads and Terms announcement, under a heading about new models rather than anything to do with retirement, is this:

"When the new models launch, all prior models will be retired."

No date. No notice period. No public name for what replaces them.

What v5 changed

v5 landed on 23 September 2025, rolling out to Pro and Premier subscribers first, with Suno calling it the world's best music model. Suno Studio launched in the same window, which was not a coincidence — the two together marked the shift from novelty generator toward something aimed at people who work with audio.

The gains were concentrated in audio quality rather than new capability. Vocals came out clearer, mixes crisper, and the muddiness that plagued v4.5 on dense arrangements largely resolved. Less common sub-genres held together better, where earlier versions tended to collapse anything unusual toward the nearest mainstream template.

If you generated heavily in 2025 and your catalogue sounds inconsistent, the v4.5-to-v5 boundary is usually where the change happened.

What v5.5 changed

v5.5 arrived on 26 March 2026 and was a different kind of release. The model improved again, but the headline was personalisation, and the three features shipped alongside it are what make the retirement notice matter.

Voices. Suno's most requested feature, available to Pro and Premier subscribers. You capture your own singing voice — live or uploaded — and generate music with it. There is a verification step that matches the voice in your audio against a random phrase you are asked to speak, and the resulting Voices are private to your account.

Custom models. You upload tracks from your own catalogue and Suno builds a personalised version of v5.5 that knows your style. Pro and Premier subscribers can create up to three.

My Taste. Available to everyone, this learns the genres and moods you keep returning to and biases generation accordingly.

Read those descriptions again with the retirement notice in mind. A custom model is a tuned version of v5.5. When v5.5 is retired, what happens to the three custom models you built on it?

The question Suno has not answered

Suno's retirement notice addresses songs. It says nothing about the things built on top of the model.

The company has been clear and reassuring about the catalogue: everything already made stays in your library, remains playable and shareable, and can still seed covers and remixes. That covers the output.

It does not cover the inputs you invested in. Custom models are tuned from a user's uploaded catalogue and are, by definition, versions of the model being retired. Voices are captured from a user's own singing and used to generate on the current model. Neither appears anywhere in the retirement announcement.

There are several plausible outcomes. Custom models and Voices might migrate to the new generation automatically. They might need retraining, which for custom models means re-uploading a catalogue and for Voices means re-recording. They might not carry over at all. Suno has not indicated which, and in fairness the next generation has not launched, so the answer may not be settled internally either.

The honest position is that this is unknown, and anyone telling you otherwise is guessing. What follows from it is simply a matter of prudence: if a Voice or a custom model is part of how you work, the finished audio you have already generated with it is the only part you currently control.

That is an argument for exporting, and it happens to land in the same nine days as the download caps.

What the next generation is

No product called v6 has been announced, whatever the search volume suggests. What Suno has described is a new generation of models built in partnership with the music industry.

The trail is easy to follow. Warner Music Group partnered with Suno in November 2025. BMG followed on 12 August 2026, announced explicitly as "part of our upcoming launch of our first music model developed with the music industry." And the v5.5 announcement in March said the capabilities it introduced — voice fidelity, personalised sound, custom models — "lay the foundation for the next generation of music models we're launching with the music industry later this year."

So: later in 2026, licensed, and per Suno's own description better on every metric the company measures, with faster generation, higher fidelity and more control.

The strategic logic is visible from the outside. Suno spent 2024 and 2025 being sued and 2026 signing partnerships, and a model trained on licensed catalogue is a different legal proposition from one trained on scraped audio. That is presumably why every prior model goes: the retirement is not a housekeeping decision about maintenance costs, it is what a clean break looks like.

Which makes the timing less mysterious than it appears. Watermarking, fingerprinting, download caps, new Terms of Service and a licensed model generation are one programme, not five announcements. Our policy tracker keeps the whole sequence dated.

Which model should you use now

For anyone generating today the answer is straightforward, because v5.5 is what you get. The more useful question is what to do about work in progress.

Finish projects on the model that started them. Regenerating a section on a different model version rarely matches, and once a version is retired that option closes permanently. If an album is half-built on v5.5, finishing it before the switch is worth more than waiting for whatever comes next.

Do not start a long project on a custom model without local copies. The custom model is the part whose future is unconfirmed. Anything it produces that you care about should exist as a file on your own disk.

Treat Voices the same way. A captured voice is quick to re-record if it comes to that, but the tracks generated with it are not reproducible if the Voice does not survive.

Do not wait for the new generation before exporting. The two events are unrelated: download caps arrive 3 September regardless of when models change, and the export allowance is the binding constraint either way.

What the model version does not change

Worth stating plainly because it drives a persistent misconception.

A newer model does not make a track more likely to pass distributor screening. Every version of every AI music generator leaves statistical artifacts in its output — spectral fingerprints, phase relationships, micro-timing distributions — and distributor classifiers read those at ingestion. v5.5 sounds considerably better than v4.5. It is not measurably less detectable, because audio quality and detectability are different properties of the file.

This catches out people who assume the rejection they got on a v4 track was a quality problem. It was not. Regenerating on v5.5 produces a better-sounding track that fails the same screen, and the same will be true of the licensed generation when it arrives — arguably more so, since a model built with industry partners is unlikely to be designed to be harder to identify.

Handling that layer is a separate step performed on the exported file. Our artifact removal ranking covers the tested options across a 50-track, six-distributor corpus.

A note on model archaeology

One quiet consequence of retiring everything at once is that Suno's back catalogue of models becomes unreproducible.

Producers who found a sound they liked on a particular version — and plenty did, with v4.5 retaining defenders for its rougher character long after v5 shipped — have no path back once a version is gone. There is no offline copy, no downgrade option, and no equivalent to keeping an old plugin installed. The model exists on Suno's servers or it does not exist at all.

This is unremarkable as a business decision and genuinely novel as a creative constraint. A guitarist's instrument does not stop existing because a better one was manufactured. A generation model does.

The practical response is the same one this site has been making since the caps were announced, for what are now three separate reasons: hold local copies of anything you consider part of your body of work. The download allowance changes on 3 September. The models retire on an unannounced date. And terms, tiers and features move continuously. None of that is unique to Suno or evidence of bad faith. It is simply what it means to have your catalogue live inside someone else's product.

The short version

v5.5 is current. v5 is what everyone still searches for. Both are going, along with every version before them, when Suno launches the licensed generation it has been building toward since the Warner partnership.

Your songs are safe and Suno has said so clearly. Your custom models and your Voices are the open question, and until the company addresses it the only part of that work you fully control is the audio you have already exported.

Nine days until the download allowance changes. The models go on their own schedule, and nobody outside Suno knows what it is.