Published 9 September 2026 · tested the day v6 launched · updated as we learn more
Suno released v6 on 9 September 2026. Three models shipped — v6 and
v6-wild for Pro and Premier subscribers, v6-mini for everyone. They were
built with Warner Music Group, BMG and Believe. The older models are already gone from the
model picker, which means every Suno guide written for v5 now describes something you
cannot select.
We tested v6 on launch day with real generations rather than reading the announcement. The field limits are unchanged — 5,000 characters of lyrics, 1,000 of style — and the one finding that surprised us is at the bottom: the difference between v6's two takes of the same generation can be larger than the effect of deliberately changing your prompt. Most prompting advice published this week is measuring that noise.
These descriptions are copied verbatim from the model picker inside the app, not paraphrased from marketing:
| Model | Who gets it | Suno's description |
|---|---|---|
v6 | Pro & Premier | “Powerful. Versatile. Refined. Our best model yet.” |
v6-wild | Pro & Premier | “Best for experimental ideas.” |
v6-mini | Everyone | “A free, more efficient version of premium v6 models.” |
There is also a Create Custom Model (Beta) option in the picker — “create a model based on your uploads”, priced at 100 credits. It is not mentioned in the launch announcement.
On our account the picker lists only the three v6 models. v5, v4.5 and v4 are not options. TechCrunch reported on launch day that Suno “plans to retire its older models” — in practice, on this account, that has already happened.
This matters more than it sounds. The overwhelming majority of Suno prompting advice online was written for v4 or v5. Metatag lists, structure tricks, workarounds for old quirks — much of it now refers to behaviour you cannot reproduce, and there is no way to A/B it against the new model because the old one is gone.
Read directly from the character counters in the v6 Advanced tab, not from documentation:
| Field | Limit | How we know |
|---|---|---|
| Lyrics | 5,000 | counter read 1534 of 5000 characters used |
| Styles | 1,000 | counter read 552/1000 |
| Exclude Styles | present, unchanged in our use | field exists in More Options |
Both are the same as v5.5. If you had a prompt workflow built around those caps, nothing needs to change.
A thing that confused us for an hour: after generating, the Styles panel shows your prompt with
extra terms appended, each prefixed with a minus — , -male lead vocal, -glossy pop-choir tone.
It looks like the model edited your input.
It didn't. We compared byte-for-byte: the style we pasted is an exact prefix of what Suno displays. Everything after it is the Exclude Styles field rendered back into the style view, with minus signs, in the order you entered it. Your prompt is stored untouched.
The practical consequence: put negatives in the Exclude field, never inline in the style box. Typed into the style box, “male lead vocal” is an instruction to include one.
Under More Options: Vocal Gender, Duration (Custom/Auto), Max Mode, Weirdness, Style Influence, Variety, and Personalize / My Taste.
Style Influence defaults to 50%. If you have written a precise style prompt, half its authority is being discarded before a note is generated. We have not yet measured the optimum setting and will not invent a number — but if your carefully specified prompt is coming back generic, that slider is the first place to look.
Same style (552 characters), same lyrics (1,534 characters), same section cues, generated hours apart on the day v6 launched — while both models still existed. That comparison cannot be repeated now that v5.5 is gone from the picker.
| v5.5 | v6 | |
|---|---|---|
| Durations returned | 2:42, 2:47 | 3:08, 3:16 |
| Style prompt storage | byte-identical | byte-identical |
| Track caption | echoes your style text | writes its own summary |
Our listening verdict, on that one track: v6 was better on vocal flow, delivery and instrumentation. That is one prompt in one genre with one voice type, judged by ear and not blind — so treat it as an observation, not a benchmark.
The clearest evidence we found. Writing its own caption for our track, v6 described “intimate chest-led Italian verses and a whispered French bridge.”
None of those three things — Italian, French, whispered — appear anywhere in the style prompt.
All three are in the section cues inside the lyrics:
[Bridge | Female — Whispered, pitch drifting slightly on the hold…].
Suno says v6 “understands more of the language and building blocks musicians use, from vocals, instrumentation and structure to mood, references and the overall feel of a song.” This is what that looks like in practice: per-section performance direction is read and used. If your lyrics are bare section headers, you are leaving that on the table.
We tried to improve one weak verse. We changed a single section cue — swapping loose, behind-the-beat phrasing for tight on-the-beat articulation — held everything else byte-identical, and regenerated.
The first take sounded worse. The second take of the same generation sounded better. Same prompt, same settings, one generation, opposite conclusions depending on which of the two takes you happened to play.
On v6, the difference between the two takes of one generation can exceed the effect of a deliberate prompt change. Which means: if you evaluate a prompt tweak by listening to one take per version — which is what almost everyone does — you are measuring noise and calling it a technique.
There is a control for this. Variety, in More Options, governs how far the two takes diverge. Turn it down when you are testing a prompt change, so the only thing moving is your edit. Turn it up when you are exploring. We found this by failing an experiment, and we are publishing the failure because the alternative — confident advice built on a single take — is how most prompting folklore gets made.
[Verse | …] brackets. Describe the delivery per section, not
only the song as a whole.Not on our account. The model picker lists only v6, v6-wild and v6-mini. Suno has said it plans to retire older models, and on our account that has already taken effect.
5,000 characters, read directly from the counter in the v6 composer. The style field is 1,000. Both are unchanged from v5.5.
Suno describes v6 as “powerful, versatile, refined”, v6-wild as “best for experimental ideas”, and v6-mini as “a free, more efficient version of premium v6 models”. v6 and v6-wild require Pro or Premier; v6-mini is available to everyone.
No. We compared byte-for-byte: the style you paste is stored exactly. What looks like an edit is your Exclude Styles field being displayed back inside the style view with minus prefixes.
Yes. In our test, v6's own description of the finished track named a “whispered French bridge” — a detail that existed only in a section cue inside the lyrics, not in the style prompt.
Take-to-take variance on v6 is substantial — in our testing it exceeded the effect of a deliberate prompt change. The Variety control in More Options governs how far the two takes diverge; lower it when you want consistency.
Suno says v6 was developed with industry partners including Warner Music Group, BMG and Believe, and trained on a new set of data that is not the same data used for previous models. Suno remains in litigation with other rights holders, so we would not describe the model as legally settled.
HookGenius writes the whole thing — the lyrics, the per-section performance direction v6 now reads, the style prompt inside the 1,000-character budget, and the Exclude Styles block that keeps your negatives out of the style box. You describe the song; you paste the result into Suno.
5 free credits · 1 credit = a full set of lyrics, a tuned style prompt and the song's structure · no card needed
HookGenius is an independent product and is not affiliated with, endorsed by, or sponsored by Suno, Inc. or any other music generation platform named on this site. Product names are trademarks of their respective owners and are used only to describe compatibility.