Suno launches v6 music-generation model

In-window event (Sep 10–17): the platform-wide v6 rollout and its launch-week consequences. Throughout the window Suno ran the v6 generation platform-wide: the three-model family (v6, v6-wild, v6-mini) was the only generation available for new creations and remakes, and prior models (v4, v4.5, v4.5+, v5, v4.5-all, v5.5) were retired from new-generation workflows — ending the era of the old free model (v4.5-all) and the v5/v5.5 paid tiers. The week's coverage analyzed and stress-tested the rollout: The Rundown (Sep 10) examined the economics of Suno's planned paid fan-remix products; independent hands-on tests were published Sep 10–17 (Undetectr Sep 10–11, piclumen Sep 11, Law Commentary Sep 11, Radio Facts Sep 11/12, tao.media Sep 12); and MusicRadar's Sep 16/17 roundup documented launch-week user backlash ("muffled, generic, soulless garbage" reports and subscription cancellations). The legal framing continued within the window: UMG/Sony's Boston case remained live with fact discovery closing Sep 30, and the first case's fair-use arguments continued to be litigated around the launch.

Antecedent context (Sep 9 launch — out-of-window, provided for completeness): On 2026-09-09 Suno announced v6, calling it "a new generation of music models developed with our industry partners, including Warner Music Group, BMG and Believe" and "our biggest step yet" (CEO Mikey Shulman). Key mechanics of the launch:

1. A three-model family — the first time Suno shipped more than one model at once: v6 (flagship; Pro/Premier; "reliable, precise… polished music across every genre and style"), v6-wild (Pro/Premier; "built for exploration," intentionally less predictable), and v6-mini (free for everyone; faster, more efficient; replacing the old free-model experience).
2. Total retirement of prior models announced — v4 through v5.5 were to be removed from new-generation workflows; the platform moves entirely onto the v6 generation. Suno had telegraphed this since the August 10 policy post and the September 5 user email ("We're retiring our older models when we launch v6"; 500 bonus credits) — the retirement itself executed during the in-window rollout.
3. Partnership groundwork — Warner settled its 2024 copyright suit on November 25, 2025, in exchange for a commitment to licensed models in 2026; BMG signed August 12, 2026; Believe and TuneCore signed September 8, 2026 — one day before launch. BMG's catalog "has begun being added," and participating Believe/TuneCore artists' music is included.
4. New creation tools — plain-language section edits, single-lyric replacement, multi-source mashups, sample/isolate/beat-building, "vibe"-based creation, and multimodal inputs; watermarking and monthly download caps (20 Pro / 60 Premier, ToS effective Sep 3) as provenance/abuse controls.
5. The legal backdrop defined the framing — the launch came nine days after Suno's September 1 court answer admitted obtaining YouTube audio via YT-DLP for training prior models, with UMG/Sony's RIAA-coordinated Boston case (June 2024) ongoing and fact discovery closing September 30.

Independent testing and user-reaction coverage across the window (The Verge's launch-day limited listening; AI Musicpreneur's 27-file controlled A/B scoring v6 74/100 vs v5.5 62/100, published Sep 9) continued to be weighed in-window (see §8).

• Before: Suno ran a single default model line (v4.5-all on free; v5/v5.5 on paid tiers) trained on undisclosed data that it now admits included YouTube audio obtained with YT-DLP. The company was in litigation with UMG and Sony (Boston, June 2024), had lost a German case to GEMA (July 31, 2026), and faced Koda (Copenhagen), SOCAN (Canada, Sep 2), Round Hill/Bright Data (Aug 17) and a Jason Isbell-led class action (Sep 1). No major-label revenue sharing existed; Warner's settlement had promised licensed models "in 2026."
• Change: Suno shipped a licensed-first, industry-partnered model family, and the platform-wide rollout plus prior-model retirement executed within the window (Sep 10–17). v6 was built, per the company, "from scratch" on licensed partner catalogs plus user preference signals, explicitly excluding UMG and Sony material; partners receive a share of revenue from launch day; older models are retired; watermarking, fingerprinting and download caps are in force; opt-in artist experiences and paid fan remixes were announced as the next phase.
• After: By the in-window rollout, the platform runs entirely on the v6 generation, with prior models (v4–v5.5) retired from new-generation workflows. Suno now has two simultaneous relationships with the industry — licensing partner of Warner/BMG/Believe and litigation defendant of UMG/Sony/others — and the business model has a licensed-revenue track. One week after the launch (Sep 18, post-window), UMG and Sony sued Suno a second time over 60,202 recordings, arguing v6 is "the fruit of the same poisoned tree" because it was trained with preference signals and (allegedly) knowledge distillation derived from the older, allegedly infringing models.

• Model architecture: Three models of different scale/intent share the v6 generation — a flagship tuned for controllable, polished output; an exploration variant (v6-wild) with higher built-in variety; and a fast, lightweight free tier (v6-mini). Suno describes improvements as coming from "R&D, the breakthroughs and the technical innovations… how these models are architected, how they are shaped, how we tune them based on preferences" (CPO Jack Brody, MBW) rather than simply from more training data.
• Training approach (company claim, unverifiable): Brody says v6 "was trained entirely from scratch, from the ground up," on "a combination of data that's been licensed from these partners, user data" — the user data being preference signals (Suno generates two takes per prompt; its record of which take users prefer informs model tuning), not uploaded user audio. Licensed repertoire is selected by the rightsholders ("Warner has basically figured out which music to hand over"). Brody states the corpus "doesn't include data from Universal or Sony." Suno's ToS (effective Sep 3) grant it a "perpetual, irrevocable" license over user Content — including outputs — for model improvement.
• Generation features: Section-level editing via natural language ("Change the chorus so it's sung by a gospel choir"); single-lyric replacement; mashups combining stems/elements from multiple songs ("Take the vocals from x, drums from y, add new lyrics… make it 80s synthwave"); sample isolation to build a beat; genre/"vibe" prompting from references; multimodal inputs (text, audio, images, video).
• Commercial rights: Paid subscribers "own the music that you generate" (company claim); outputs are "considered an original creation"; Suno assigns rights in outputs to Pro/Premier subscribers subject to commercial-use restrictions, with no warranty that copyright vests in any Output. Future derivative (artist-opt-in remix) products will have different rights terms.
• Licensing economics: Partners receive "a portion of our revenue" shared "from launch day," with rightsholders deciding distribution to artists (company claim; amounts undisclosed). Downloads are capped (20/month Pro, 60/month Premier; extra purchases; unlimited Studio exports on Premier) to curb mass export/streaming-fraud, with watermarking/fingerprinting commitments.
• Retirement mechanics: Prior models were deactivated for new generations as v6 rolled out; the model picker now shows only v6, v6-wild, v6-mini.

• First major generator built with the industry, not just sued by it: v6 is the first Suno model family developed with record-label participation after the November 2025 Warner settlement — a structural shift from "train first, litigate later" to licensed-partner development, and the clearest real-world test of whether licensed training data sustains label-quality output.
• The copyright fight did not pause for the launch: UMG and Sony remain plaintiffs; Suno admitted YT-DLP-based YouTube sourcing on Sep 1; fact discovery closes Sep 30; GEMA already won in Munich. The launch is as much a legal-strategy reset as a product event — and it provoked a second UMG/Sony suit (Sep 18, post-window) arguing v6 merely laundered prior infringement via preference signals and knowledge distillation.
• Pushes the boundary of label-quality generative music: independent tests (The Verge; AI Musicpreneur's 27-file A/B: v6 74/100 vs v5.5 62/100) show real gains in genre understanding, vocal control and mix polish — with full 2:30–3:41 songs generated in 10–20 seconds — while launch-week user backlash (MusicRadar: "muffled, generic, soulless garbage"; "fidelity up, feel down") shows the quality ceiling is contested.
• Creates the licensed-AI-music template the labels chose to fight over: Suno's own deals become evidence in the labels' argument that "a functioning market exists for licensing sound recordings to train generative AI models" — a double-edged sword for the company's fair-use defense in the first case.

• Licensed generative music at consumer scale: users can now generate, in seconds, full songs from a model trained on licensed catalogs (Warner, BMG, Believe/TuneCore) with revenue flowing to rightsholders — the first mainstream instantiation of the "licensed training data" business model in music.
• Artist opt-in products and paid fan remixes: Suno announced phase-two products where artists choose to participate and get paid, and fans pay to personalize/remix music from participating artists — with a distribution path to DSPs via Believe/TuneCore. Consumer pricing, launch timing and artist pay formulas are still open.
• Granular post-generation control: plain-language section and lyric editing, mashups and sample-based beat-building convert a text-to-song tool into an editable music-production surface — closer to a DAW than v5 was (complementing Suno Studio 2.0, launched Aug 13, 2026).
• Provenance-led distribution: watermarking + fingerprinting + download caps give distributors and platforms (Deezer, Spotify) mechanisms to govern AI-content flood — Deezer reported ~90,000 AI-generated tracks per day (>50% of daily new arrivals) in July 2026, the scale problem these controls target.

• Resolved adjudication note: per orchestrator adjudication (user directive), the story's event is re-anchored to the in-window platform-wide v6 rollout and coverage (Sep 10–17); the Sep 9 launch announcement is documented as antecedent context with its facts, dates and sourcing unchanged.
• What exactly is in v6's training set — which catalogs, how much material, and can "preference signals" from two-take generations of older models be cleanly excluded from a "from-scratch" model (the factual core of the second UMG/Sony suit)?
• Can UMG/Sony prove knowledge distillation from prior models into v6, and will Judge Saylor's first-case fair-use approach lead to an early ruling that constrains the second case?
• What are the actual revenue-share percentages, the artist pass-through, consumer pricing and launch date for paid fan remixes — and how many artists will opt in?
• Will launch-week quality complaints trigger model refinements (v6.x) within weeks, as past Suno releases have?
• Do watermarking + fingerprinting + download caps measurably reduce AI-track flood on DSPs (Deezer's ~90k/day baseline)?
• More partner deals? Brody: "That is absolutely our goal" — which rightsholders (Koda, SOCAN, Merlin, independents) convert from plaintiff to partner, and does the second suit freeze that pipeline?

• Immediate (≤2 weeks): Suno's first-case fact discovery closes September 30; the second UMG/Sony suit (filed Sep 18, post-window) proceeds in Boston under Judge Saylor's docket; Suno's response to the "poisoned tree"/distillation allegations will be filed within weeks — watch for counterarguments on preference-signal licensing terms and technical separation.
• Near term (Q4 2026): opt-in artist experiences and paid fan remixes — launch date, consumer pricing and artist-payout formula expected; more rightsholder deals (Brody: "absolutely our goal"); watch whether Koda/SOCAN settlements follow the Warner/BMG pattern; watermark rollout; a probable v6.x refinement responding to documented mixing complaints.
• Medium term: Judge Saylor's fair-use ruling in the first case (after Sep 30 discovery) will shape both suits and the wider "training on copyrighted works" doctrine; the GEMA precedent will move through European appeals while the EU AI Act's copyright-related GPAI obligations begin applying; Spotify's fan-remix tool (UMG + Merlin) comes to market against Suno's remix plans — the paid-remix category becomes a live competitive arena.
• Signal to track: any published disclosure of v6 training composition, the first artist-opt-in count for remix products, and the labels' first concrete damages figure in the second case — the three numbers that will define whether the licensed pivot is a settlement engine or a litigation accelerant.

The story of the week in generative music is not "better AI songs" — it is Suno shipping its first model family built with the record industry, retiring every model it was sued over, and still getting sued again nine days later. The clean hook: Suno replaced its models with Warner, BMG and Believe on board, cut Universal and Sony out of the training data — and the labels it excluded sued it over the models it retired, arguing the "clean" v6 is "the fruit of the same poisoned tree." The durable angle is the precedent: a mainstream generative-AI company publicly converting from unlicensed to licensed training, showing both that the licensing market works and that it cannot erase what came before — a template every generative-AI company (music, video, image, text) will now be measured against by plaintiffs, partners and regulators alike. (Note: the September 9 launch is antecedent context; the story's in-window event is the September 10–17 platform-wide rollout and its launch-week consequences, per orchestrator adjudication.)