Suno Launches v6: First Licensed AI Music Models Built on Warner Music, BMG, and Believe

Suno retired its entire previous model lineup on September 9 and replaced it with v6, the first generation of AI music models the Cambridge, Massachusetts company says was built using catalogs licensed from Warner Music Group, BMG, and digital distributor Believe — with revenue sharing to those three partners beginning the same day. The move is the most concrete step any AI music platform has taken toward a model where training rights and commercial royalties are negotiated simultaneously rather than resolved in court after the fact. It is also, in a legal irony the company's opponents are already exploiting, evidence that a licensing market for AI training data exists — the precise argument Universal Music Group and Sony Music are pressing in the ongoing federal lawsuit that v6 did nothing to settle.

Three Models, One Platform Reset

The v6 family consists of three distinct models differentiated by creative posture rather than capability tier alone. v6 is the flagship, available to Pro and Premier subscribers, and is positioned for consistent, genre-faithful results. v6-wild, also on paid tiers, is described by the company as deliberately "less predictable and more varied" — built for experimentation where the unexpected is the point. v6-mini is free to all users and, according to CPO Jack Brody, faster than its predecessors — delivering finished tracks within seconds of a prompt.

The launch retired every prior Suno generation. Users retain access to songs made under v5 or v5.5, but the models that produced them are gone from the platform. That transition has a precedent embedded in it: when Suno settled Warner Music Group's copyright lawsuit in November 2025, the company pledged that its current models would be deprecated when properly licensed successors arrived. The v6 launch makes good on that promise.

Brody said in an interview with Music Business Worldwide that "v6 was trained entirely from scratch, from the ground up" on data that "doesn't include data from Universal or Sony." That exclusion is both a legal reality and a commercial constraint: Universal Music and Sony Music remain active plaintiffs in a federal case in Boston, and Suno's decision to keep their catalogs out of v6 training is partly an attempt to limit the scope of what those labels can claim in discovery.

How Suno's Models Actually Generate Music

Suno has not published a technical paper for any of its model generations, and v6 is no exception. What is known about its architecture comes from a combination of patent filings, third-party analysis, and statements from Brody and CEO Mikey Shulman. The broad picture: Suno uses a latent diffusion transformer architecture, a hybrid approach that encodes audio into a compressed latent representation, then applies a transformer-based reverse diffusion process to generate output audio from that space.

The key architectural choice that differentiates Suno from many text-to-audio research models is that it generates audio tokens sequentially — more like a language model generating words one after another than the parallel denoising process used in image diffusion. That approach is slower per token but significantly better at preserving musical structure across a full song, where a chorus needs to recall the melodic patterns established in a verse three minutes earlier.

Brody told Music Business Worldwide that v6's improvements come less from the licensed catalog itself than from accumulated engineering research: "A lot of the things that make these models better are actually not the data it is trained on. It is the R&D, the breakthroughs and the technical innovations that go into how these models are architected, how they are shaped, how we tune them based on preferences, and all of that."

That tuning refers to reinforcement learning from user preferences built into Suno's platform from the start. Every creation generates two versions; the user's selection — which they keep, which they discard — functions as a training signal. Brody confirmed that v6 incorporates years of this preference data from Suno's 100 million total users alongside the licensed catalog. The training corpus is not simply Warner's or BMG's tracks; it combines those with a large body of accumulated signal about what generated music users actually prefer.

What v6 Can Do That Prior Suno Models Could Not

The generative capabilities in v6 represent a meaningful functional expansion beyond the single-prompt-to-full-song workflow that defined earlier versions. The most practically significant is section-level editing in natural language: a user can now ask the model to change only the chorus while leaving the verse untouched, or swap a single lyric without regenerating the rest of the track. Suno's documentation illustrates this with an example asking the model to make a chorus "sung by a gospel choir" — the arrangement around it stays, only the treatment of that section changes.

The mashup capability combines elements from multiple sources in a single generation request. A user can specify vocals from one existing track, drums from another, a new lyric theme, and a style direction — and receive a unified output synthesizing all four instructions. This is not remix technology in the traditional sense; it is a multimodal understanding of the user's creative intent that the model interprets as a composition brief.

A third new workflow handles sampling at timestamp resolution: "Sample the riff at 0:45, isolate the guitar, build a beat around it." The model identifies the target moment, separates the instrument from the mix, and treats the isolated audio as the foundation for a new generation. That combination — timestamp reference, source separation, and generation — has required three separate tools in conventional audio production workflows.

Input is now multimodal as well. v5.5 accepted text and audio. v6 accepts text, audio, images, and video. A user can supply a phone video of an environment or a concert, or a photograph, and request music that fits the visual material. That positions Suno in the same input territory as video-first AI tools rather than purely text-to-song generators, expanding the range of creative workflows it can replace or augment.

The Licensing Architecture and What It Does and Does Not Cover

The partnership model that underpins v6 is meaningfully more complex than "Suno licensed some songs." The three deals are structurally different from one another, and none covers the full catalog of any partner.

The WMG deal, finalized in November 2025 as a settlement of Warner's copyright lawsuit, gave Warner input into which catalog tracks entered training. Brody told Music Business Worldwide: "Warner has basically figured out which music to hand over as part of that process for training." The specific tracks and the financial terms of the WMG settlement are not public. As part of the deal, Suno also acquired Warner's live-music discovery platform Songkick.

The BMG deal, announced August 12, is structured differently. BMG was never a plaintiff against Suno, so there was no lawsuit to settle — the agreement instead retroactively compensates BMG for Suno's prior catalog use and licenses BMG's repertoire going forward. The deal is opt-in: only artists who actively elect to participate have their music licensed to Suno. BMG's roster includes publishing rights to Mick Jagger, Keith Richards, Bruno Mars, and Diane Warren, among others; how many choose to opt in is a question the deal leaves to the artists themselves.

The Believe deal, signed September 8, reversed a stance the company had taken just five months earlier — in April 2026, Believe had blocked distribution of all Suno-generated tracks through its platform and TuneCore. The new agreement, also opt-in, restores distribution access for artists who consent.

Revenue sharing begins immediately, with Suno distributing a portion of its platform revenue to the three partners. The partners are then responsible for distributing those payments to their artists. Neither the formula for calculating Suno's payment to partners nor the downstream formula for artist payments has been disclosed publicly.

What the licensing architecture does not cover is as important as what it does. UMG and Sony are explicitly excluded from the v6 training corpus. That means music from artists whose master recordings are controlled by Universal or Sony — which together represent a substantial share of commercially popular music globally — played no role in training the models that now power Suno for 100 million users. Round Hill Music, which manages a catalog of independent evergreen compositions and refuses to settle, is also unaddressed. The scope of "licensed" in v6 is the scope of three specific commercial agreements, not a broad industry consensus.

Where Suno Sits in the AI Music Competitive Landscape

The licensing development shifts Suno's competitive position relative to Udio, its primary rival for full-song vocal-capable generation, in a way that is not straightforward.

Udio settled its UMG copyright case in October 2025 and followed with deals covering Warner, Merlin, and Kobalt — giving it, at that point, more comprehensive licensing coverage than Suno. The tradeoff was the "walled garden" model that UMG has pushed as its preferred architecture for AI music: as a condition of its partnership changes, Udio disabled all user audio, video, and stem downloads. As of September 2026, you can create music on Udio but cannot export it to distribute elsewhere.

Suno has consistently resisted that model. The company's chief music officer Paul Sinclair argued in February for "open studios, not walled gardens," and the v6 launch maintains that position: paid subscribers can download and commercially distribute what they create within the platform's terms. That distribution openness has been a meaningful differentiator for creators who need to place music on streaming platforms, in YouTube videos, or in commercial work.

v6 gives Suno WMG and BMG coverage — partially overlapping with Udio's licensing — while keeping downloads available. The cost is that Sony and UMG catalog is explicitly absent from v6 training, while Udio's UMG deal covers Universal's artists (though Udio's outputs cannot leave the platform). For enterprise users or sync licensing professionals, neither platform currently offers a clean solution, but they offer different incomplete solutions.

ElevenLabs Music, which launched in August 2025 with licensing in place from the start, has a cleaner training data history but less capability for complex, genre-wide vocal song generation. Stable Audio 3 from Stability AI, which launched in May 2026 with open weights that run on consumer hardware, has no vocal generation and does not compete for the same use case.

Read more: Google Lyria 3.5 enters the AI music market as a licensed rival

The Legal Exposure v6 Does Not Resolve

The v6 launch addresses Suno's legal exposure for its forward-going operations with three partners. It does not address the core legal questions that the UMG/Sony lawsuit was designed to force into a courtroom.

The lawsuit in the US District Court for the District of Massachusetts alleges that Suno's earlier models — the ones v6 just replaced — were trained on copyrighted recordings without authorization. On August 25, UMG and Sony filed an amended complaint adding a stream-ripping claim, alleging that Suno bypassed YouTube's anti-download technology to harvest audio for training. Suno answered on September 1, admitting in its filing that "audio data was obtained from YouTube for use as training data using YT-DLP" — the open-source download utility — while arguing the labels lack standing to bring the stream-ripping claim under the Digital Millennium Copyright Act. Fact discovery in that case closes September 30. Dispositive motions are due April 9, 2027.

There is a compounding irony in Suno's legal position. By signing licensing agreements with WMG and BMG, Suno has demonstrated that a commercial market for licensing AI training data exists. UMG and Sony are now using that fact as evidence against Suno's fair-use defense: under US copyright law, fair-use analysis considers whether the defendant's use could substitute for a legitimate licensing market. If such a market demonstrably exists — as the WMG and BMG deals confirm — it becomes harder to argue that prior training without a license was fair. Suno's agreements are simultaneously its clearest signal of good faith and the strongest evidence its opponents have that the earlier use was compensable.

Outside the United States, the Munich Regional Court ruled on July 31 that Suno had infringed GEMA-represented composers' rights in training its earlier models, ordering the company to stop unauthorized reproduction, disclose the scale of use, and pay damages — a ruling immediately enforceable while Suno evaluates an appeal. A referral before the European Court of Justice (Case C-250/25) could extend comparable obligations across the EU.

Round Hill Music's August 17 lawsuit adds another front the company refuses to negotiate away. Every major settlement in AI music copyright litigation has occurred before any court issued a final binding ruling on the fair-use question. Round Hill's stated commitment to a jury verdict, combined with its choice of attorney Richard Busch — who won the landmark Blurred Lines infringement case — signals an intent to produce the first binding precedent rather than trade the legal threat for a licensing deal. Jason Isbell and three other artists filed a separate class action on September 1 alleging Suno used names and likenesses without consent. SOCAN filed in Canadian federal court on September 2.

A Phase 2 on the Horizon, and a Ruling That Will Reshape the Industry

Shulman framed the v6 launch as the first of two phases. The second, described in broad terms, involves "opt-in experiences built around individual artists, where artists can choose to participate and get paid when they do, and fans get new ways to connect with the artists they love." No timeline beyond "some point in the coming months" was offered.

A "Create Custom Model" feature, visible in Suno's platform but not included in the official launch post, is labeled Beta and priced at 100 credits — letting users train a personalized model on their own uploaded tracks, an early iteration of what the artist opt-in experience could eventually look like at scale.

The marker that matters most for the AI music industry is the US federal court calendar rather than Suno's product roadmap. The April 2027 dispositive motion deadline in the Massachusetts case will bring the fair-use argument to a head — not just for Suno, but for any AI company that trained on copyrighted material before licensing agreements existed. If the court rules that prior training was not fair use, Suno's licensing architecture becomes the mandatory industry floor rather than a competitive choice. If it rules the prior use was fair, the WMG and BMG deals look like voluntary overpayment. Either outcome will define what AI creative companies owe the industries they learned from.