Suno V6 Deep Dive: One Launch, Three Models, and the Fine Print on Stems

Three Boxes, One Ship Date

When we reviewed V5 in July, the naming problem was stacking: V5, then V5.5, then Persona bolted on top, all answering to "Suno V5." The V6 problem is breadth. One ship date, three boxes, and the split reads less like a product ladder than a bet on three different listeners.

| Tier | Stated role | Price, per this snapshot | |------|-------------|--------------------------| | v6 | Flagship; controllable, consistency-first | Not published in the coverage cited here | | v6-wild | The exploration tier | Not published in the coverage cited here | | v6-mini | Free, lightweight entry | Free |

The flagship's brief is repeatability. Consistency, in a music model, means the same brief yields the same character of output across cuts, which is the property you care about when a track is an asset instead of a one-off: a brand jingle that needs a dozen variants, a series intro that has to sound like itself every episode. Controllable is the vendor's word for prompt obedience. I have no independent measurement of it from this snapshot, only the launch framing, plus a July corpus that showed the V5 generation holding verse-chorus-bridge structure across a full song.

v6-wild pitches the same engine in the opposite direction: variance as a feature. You reach for it while hunting, in ideation, when a wrong take that sparks a right one is worth more than a reliable copy of your last track. That is close to the entire published description. How the variance is dialed, whether it trades off against the flagship's consistency controls, what it costs: none of that appears in the coverage this review draws from.

v6-mini keeps a door open for the listener who will never pay. Free and lightweight continues the freemium pattern Suno ran through the V5 cycle, where the free tier bought kicks and the paid tiers bought shipping rights. What mini's limits are, though: daily cap, quality gap against the flagship, download rights. Three unknowns, none resolved in this snapshot, and those three unknowns decide whether mini is a demo or a habit.

Three boxes, one engine.

The Edit Features Are the Actual Product

Tier splits are packaging. The capability list is where this launch changes work, and two of the four items change it more than the other two combined.

Natural-language local editing means describing the change in a sentence and having the model apply it to the section you named. The old unit of edit was the whole song: you rerolled and hoped the hook that made you pick this take survived the dice. A chorus you can ask, in words, to calm down converts regeneration from a gamble into a revision. Anyone who has burned an evening rerolling a track that was ninety percent right already understands the size of that shift.

Single-line lyric editing is the same idea at finer grain. The lyric sheet used to be the smallest editable unit; you fixed one clunky line and reran the vocal, hoping nothing else moved. Line-level edits make words behave like text again, which is to say fixable.

Multimodal input and the compliance hardening round out the announcement, and both arrive underdescribed. The launch coverage names each without enumerating which modalities are accepted or which guardrails changed, so this review will not guess at either. What the compliance work plausibly responds to is no mystery: Suno spent 2026 settling with rightsholders, and our July review covered the Warner settlement and its squeeze on the free tier. A vendor with that history hardening its input checks is a reasonable reading, and I am marking it as a reading until somebody documents the boundaries.

Stated as cost, before against after: last cycle, one note of dissent meant a full regeneration and a fresh gamble on every section. This cycle, per the September 9 list, the fix is scoped to the line or the section you name.

Stem Separation, With an Asterisk

Stem separation splits a finished track into components: the vocal on one stem, the drums on another, the rest of the arrangement on a third. It is the difference between receiving a song and receiving the raw material of a song, and it is the item on this list with the widest gap between how it reads in a launch post and what it obligates anyone to deliver.

Here's the trap. A feature list travels with a product, not with a model. Stem separation on the V6 list is a capability of Suno's own product surface: open Suno, generate, split. Nothing in the launch coverage says the feature follows the engine into third-party integrations, because capabilities are product decisions first and model properties second, if they are model properties at all.

That distinction is usually pedantry. This quarter it decides real money, because Suno V6 now runs somewhere that does not ship stems at all.

Running on x-rush, Stems Not Included

Since my employer enters the frame here, the straight version: Music Gen on x-rush.ai runs on Suno V6, and the frozen workflow that governs it delivers something narrower than Suno's own feature page. The narrowing is deliberate, and every difference is checkable against the workflow's own documentation.

What Music Gen delivers. Two modes, one per workspace: a Song comes out with a lead vocal, Pure Music comes out without one and works as standalone listening or BGM. Both bill the same, one complete track per generation at 5 credits, with the price on the confirm button before anything runs. A Workflow Copilot sits in every workspace and can draft lyrics or complete your genre, mood, and structure directions; its successful assists bill through an account-wide pool at 0.5 credits per 10 assists. A track ships only after delivery checks pass, and it ships as MP3 and WAV of the complete audio. A task that cannot deliver puts its reserved credits back. Parallel workspaces follow your plan: 1 on Free, 2 on Lite, 3 on Pro. And requests that ask to reproduce an existing song, cover a melody, or imitate a recognizable singer stop before any credit is reserved.

What Music Gen does not deliver, just as plainly: no stems or multitrack output, no vocal/instrumental split, no vocals-only export, no section-level redo. One generation is one complete track, and changing a verse means a new 5-credit unit. No reference audio, no covers, no voice cloning, no MIDI input.

So the asterisk resolves like this. Suno's stem separation is real, and it lives in Suno's product. Music Gen's first version does not deliver stem separation, does not deliver a vocal/instrumental split, and no page on this site should imply otherwise. If what you need is an a cappella and an instrumental off the same finished track, that is not on the menu here today; the honest routes are Suno's own product or a dedicated separation tool. I would rather write that sentence than let a feature list drift across a platform boundary.

The same honesty cuts the other way, at the editing unit. Suno's single-line lyric editing is a Suno-product capability. Inside Music Gen there is no in-tool editing at all, so the remedy for a wrong note is an edited input and a fresh 5-credit generation. You trade Suno's product surface for a flat price shown up front, delivery-checked files, and credits returned on failure. Which trade wins depends on whether you need the studio or the delivery.

What the Coverage Leaves Out

• Per-tier pricing for v6 and v6-wild: absent from the coverage cited here.
• Independent rankings or Elo for any V6 tier: none in this snapshot. The July boards (V5.5 at 1,293 Elo, first on both Artificial Analysis music boards, per our V5 review) predate the family and say nothing about it.
• v6-mini's limits: cap, quality gap, and download rights, all unpublished.
• The modalities behind multimodal input: unnamed.
• The mechanics of the compliance hardening: unnamed.
• Whether the edit features or stem separation travel with the engine into integrations: unstated, and the Music Gen section above is evidence that at least one capability does not.

None of that is a knock on the models; it is a statement about the evidence. Every claim in this review is tethered to two dates, September 9 for the family and October 12 for this snapshot, and the missing items are all checkable later. The review date of December 20 exists to check them.

Who Picks Which Box

Use v6 if your output is an asset with a brand on it and this month's cut has to match last quarter's. Use v6-wild while you are hunting, before you know what you want. Use v6-mini to learn the interface at zero cost, which is the role the V5-era free tier filled. Skip the platform assumption entirely if stems are your deliverable: Suno's own product splits them, Music Gen does not, and no amount of feature-list optimism moves a frozen workflow.

Beyond that, the decision is pricing-shaped, and the pricing sits on the missing list. Re-check before you commit a budget to any tier.

Recap, in numbers: one launch on September 9, 2026, split three ways. Four headline capabilities, two of which (local editing, line-level lyric edits) change the unit of work. One capability, stems, that belongs to the product and not the pipeline. On the platform side: 5 credits a track, 0.5 per 10 copilot assists, MP3 plus WAV, credits back on failure.

Pick the box that matches your tolerance for surprise.

Or skip the tier math and describe the song you need, meeting the price on the confirm screen instead of in a spec hunt.