Suno v6 arrives with Warner, BMG and Believe: music AI models trained under license, old models retired, labels paid

Suno v6 arrives with Warner, BMG and Believe: AI music models trained on licensed data, old models retired, labels paid

🔎 The day AI music left the courtroom

On September 9, 2026, Suno launched its v6 suite: three music generation models developed with Warner Music Group, BMG and Believe, and trained "entirely from scratch, from the ground up" on licensed data. The same day, the company retired all of its older models — precisely the ones that had landed it the most high-profile lawsuits in generative AI.

The timing is no coincidence. In November 2025, Suno settled its lawsuit with Warner by promising "new, more advanced, licensed models" for 2026. Ten months later, the promise has been kept: Music Business Worldwide confirms that v6 is built on the licensed catalogs of the three partners, supplemented by user data and the team's R&D learnings. Digital Music News sums it up the same way: the "influence first, negotiate later" bet has just flipped into the "license, then share" bet.

Why does this matter now? Because Suno is no longer a marginal startup. More than 100 million users, more than 2 million paying subscribers, $300M in ARR in early 2026, a $5.4B valuation: at that scale, legal ambiguity was no longer sustainable. v6 is the first real-world, full-scale test of the "licensed training + revenue sharing" pairing applied to generative AI — and the entire industry, music first and foremost but not only, is watching to see how it holds up.

Key takeaways

• On September 9, 2026, Suno launched three v6 models developed with Warner Music Group, BMG, and Believe: v6 (flagship, reserved for Pro/Premier subscribers), v6-wild (ideation, paid subscribers) and v6-mini (free).
• All previous models were removed that same day, according to CPO Jack Brody — a move as symbolic as it is legal.
• v6 was trained from scratch on data licensed from partners, user data, and R&D learnings. No recordings from Universal Music Group or Sony Music in the training set.
• A share of Suno's revenue is split with the labels from launch; Warner artists benefit from it, and Believe/TuneCore artists can opt in.
• The launch comes with watermarking, fingerprinting, and monthly download caps depending on the plan.
• Gray areas remain: exact catalogs not disclosed, the status of training data predating the agreements, and public silence from UMG and Sony.

Recommended tools

This launch is already featured in our new AI tools watchlist — and it's likely to set a precedent well beyond music.

Three models, three use cases: what Suno v6 really changes

Suno v6 isn't one model — it's a suite of three, designed to separate exploration from production. That's the first takeaway from this launch: Suno is professionalizing its own architecture.

v6 flagship is the "results" model: reliable, precise, reserved for Pro and Premier subscribers. It's the one meant to produce tracks that are usable as-is.

v6-wild, also reserved for paying subscribers, embraces the opposite: deliberately unpredictable ideation, to be refined later in the flagship. Suno is thus structuring a genuine studio workflow — explore broadly, then narrow down. It's a creative logic, not a slot machine.

v6-mini is free, and Suno claims it's faster than the platform's other free models. Free access remains capped by limits, but it shows that freemium isn't doomed to be bad — we actually covered this for LLMs in our guide on using free models without sacrificing quality.

Feature-wise, The Decoder and Quartz confirm the leap: precise section editing in natural language — changing a verse without rebuilding the song —, editing a single line of lyrics without regeneration, multi-source mashups in a single request, and multimodal generation from text, audio, images, or video.

My take: it's partial editing that changes everything, not generation. Being able to touch up a verse the way you'd edit a paragraph turns Suno into an iteration tool — and therefore a real work tool — and durably brings AI music closer to producers' actual practices.

Why Suno removed its old models the very same day

Because those old models embodied exactly the legal risk that v6 is meant to erase. Keeping them in production meant keeping online the very objects of the lawsuits.

CPO Jack Brody confirmed it to Music Business Worldwide: the old models were pulled the very day v6 launched, trained "entirely from scratch" on a radically different dataset — licensed partner data, user data, and the team's R&D learnings.

The decision serves a dual purpose. Product first: a single, coherent catalog, on which watermarking and fingerprinting apply end to end. Legal second: the models under attack vanish from circulation, and the message to partners as well as to the courts is crystal clear — the page is turned.

Important nuance: erasing a product settles nothing about the past. The question of the data used before the agreements remains entirely open, and we'll come back to it — it's probably the real legal battleground of 2027.

Warner, BMG, Believe: the revenue-sharing mechanics

The arrangement came together in three stages. November 2025: settlement with Warner and a public promise of licensed models for 2026. August 12, 2026: agreement with BMG, with artist opt-in and a settlement explicitly covering prior use of the BMG catalog. September 8: Believe and TuneCore join on the eve of the launch. The full timeline is summarized by MusicTimes.

On the money, Brody put it bluntly to TAO Media: "a portion of our revenue will be shared with our partners," starting on launch day. The percentage, the eligible repertoire, and the accounting mechanics, however, are not public. That's the price of a deal negotiated behind closed doors.

At Warner, the tone is set by CEO Robert Kyncl, quoted by ecoustics: the label's artists and songwriters will share in the revenue generated by Suno's more than 2 million paying subscribers. This is an unprecedented "creation-based" stream — the money comes from the use of AI tools, not from streaming finished recordings — and Warner expects a material contribution as early as its fiscal year 2027.

Two more entry points complete the setup. Believe/TuneCore artists can opt in and get paid for approved uses, and tracks created with the partner model will be distributable via Believe and TuneCore. Above all, an opt-in program is in development: individual artists will be able to take part in the AI remix features and be paid when their music is used, according to Quartz.

You have to grasp what Suno can afford to share. 2 million paying subscribers and $300M in ARR at the end of February 2026, versus $200M three months earlier, according to TechCrunch; a $400M Series D at a $5.4B valuation in June 2026; more than $819M raised in total, according to PitchBook. Suno is joining the very exclusive club of consumer AI products at mass scale — much like when Meta turned OpenClaw into a mass-market product with its personal agent.

The comparison with the other giants is instructive. When OpenAI rolls out ChatGPT Ads with WPP and Omnicom, consumer AI is funded by advertising. Suno, for its part, chooses to share with rights holders. Three models now coexist to fund AI: subscription, ads, royalties. The third has just come into its own.

What Suno Doesn't Say About Its Training Data

Suno communicates extensively about its partners, and almost never about its data. That's the whole problem with this launch, which was otherwise remarkably well executed.

The Decoder points this out: Suno describes Warner, BMG, and Believe as "codevelopers," without detailing either the catalogs used or the volumes involved. "Licensed with Warner" and "trained on X thousand hours of the Warner catalog" are not the same piece of information, and only the first is public.

The most awkward legal nuance comes from Pitchfork, as recalled by MusicTimes: from the Warner deal onward, the label spoke of "licensed" models without ever saying "fully licensed." The BMG settlement, which explicitly covers prior use of the catalog, implicitly confirms that this prior use existed — and it doesn't say at what volume.

What is asserted, on the other hand, is the absence: v6's training set includes no recordings from Universal Music Group or Sony Music, according to Brody. The two majors have not publicly reacted to their exclusion from the launch, and their lawsuits against Suno are still ongoing — The Decoder sums up the situation with a cruel "Warner caves first, Universal and Sony keep suing."

My analysis: watermarking and fingerprinting make downstream usage traceable, but upstream — the training — remains opaque as long as UMG and Sony haven't settled. "Licensed" is, for now, a partial promise, and Suno has every interest in turning "partial" into "total" before the question is decided in court.

Suno vs Udio: two opposing visions of AI music

Two legal settlements, two incompatible product architectures. This is the real fault line in the AI music market as of September 2026.

Udio signed with Universal, Warner, Merlin, and Kobalt — the broadest legal coverage on the market — but within a "walled garden" with restricted downloads. Suno explicitly rejects this model, according to Brody: downloads remain possible, but with monthly caps per plan, presented as a brake on mass exports and streaming fraud.

Neither approach is "open". Suno locks down quantity, Udio locks down output. Suno's choice is riskier — it keeps Suno in the crosshairs of streaming platforms, which view mass uploads of generated content very unfavorably — but it respects the platform's founding habit: create at home, publish afterward.

Mixed reception: artists remain divided, users lose their bearings

v6 is hailed as a legal turning point, but Suno's legitimacy remains fiercely contested. The reception is anything but a unanimous triumph.

On the artists' side, the distrust has not evaporated with the deals. The open letter "Say No to Suno", launched by artist rights advocacy groups, and the openly declared opposition of figures such as Billie Eilish, Chappell Roan, or Katy Perry, documented by TechCrunch, continue to carry weight. The absence of UMG and Sony from the launch says as much as Warner's presence.

On the users' side, the removal of the old models on the very same day comes at a real and immediate cost. The prompts, styles, and sonic references built on the old models don't automatically transfer to v6, which was trained from scratch. Those who had made the distinctive "touch" of the older versions their creative identity no longer have a choice: their go-to tool vanished overnight.

This is the price of going legit, and Suno owns it: better a few months of frustration than a courtroom future. The bet is rational — a short-term net loss traded against long-term viability — but we shouldn't pretend the frustration doesn't exist.

"Lawsuit, license, revenue share": the new standard for generative AI?

In music, yes: the pattern is already a de facto standard. Elsewhere — image, video — it's the model everyone is studying, without yet daring to replicate it exactly.

Udio followed exactly the same path as Suno: lawsuit, settlement, license, revenue sharing, with UMG and WMG to boot. As MusicTimes sums up, the sequence has become the norm in AI music in under a year. What looked like an isolated capitulation by Warner in November 2025 has become doctrine.

Why does it work? Because the money is there on both sides. Suno can pay — $300M ARR, $5.4B valuation — and the labels gain a new revenue stream, backed by millions of subscribers rather than streams. Artists, finally, get an entry point through opt-in. On paper, nobody loses — except perhaps the more troubling question of creative diversity in the era of catalogs licensed by default.

Fortunately, there is a counter-model, and it doesn't go through the courts: open source. While Suno locks its data behind confidential agreements, DeepSeek V3.1 ships under the MIT license and iFlytek opens up its edge models Spark X2.5 with 1M tokens of context. Two philosophies are clashing: pay for the right to train, or publish the weights and own the debate over the data. The music industry has come down on the side of the former; the rest of AI isn't done arbitrating yet.

My prediction: music was the ideal laboratory — concentrated rights, identifiable rights holders, technically mature watermarking. Image and video, with their fragmented rights chains, will replicate the model more slowly. But by the end of 2027, every major consumer-facing generative product will have either a licensing deal or a lawsuit in progress. Often both, in succession.

❌ Common Mistakes

Mistake 1: believing that "licensed" means "fully clean"

The word "licensed" was never accompanied by "fully" in official announcements, as Pitchfork pointed out back when the Warner deal was announced. The status of training data used before the deals remains a gray area. The fix: read press releases word for word and follow UMG's and Sony's procedures — they are the ones who will legally settle the question.

Mistake 2: ignoring watermarking, fingerprinting, and download caps

v6 ships with audio watermarking, acoustic fingerprinting, and metadata controls tied to downloads. Mass exports to streaming platforms are the stated target of these measures — Brody explicitly talks about streaming fraud. The fix: go through official distribution (Believe/TuneCore for partner-model tracks) rather than automated bulk uploads.

Mistake 3: forgetting that what you create feeds Suno

Suno's terms of service retain a "perpetual, irrevocable" license on all user content to improve its models — and v6's dataset explicitly includes user data. The fix: if you're working with sensitive references (identifiable voices, unreleased material), think twice before uploading them as a source.

Mistake 4: confusing a paid subscription with commercial rights

Being a Pro or Premier subscriber gives you access to the v6 flagship, not universal commercial carte blanche: download caps per plan, systematic watermarking, and distribution restricted to partner channels. The fix: check the terms of your plan before promising commercial exploitation to a client.

❓ Frequently Asked Questions

Is Suno v6 free?

Partially. v6-mini is free and presented by Suno as faster than the platform's other free models. The flagship v6 and v6-wild are reserved for paying subscribers. Note: monthly download caps apply to all plans, including paid ones.

How much do the Pro and Premier plans cost?

According to Quartz, Pro costs $8/month and Premier $24/month (US pricing, September 2026 — check on suno.com). Both give access to flagship v6 and v6-wild. These are precisely Suno's roughly 2 million paying subscribers whose revenue now funds the revenue sharing with the labels.

Can I distribute my Suno v6 tracks on streaming platforms?

Tracks created with the partner model will be distributable via Believe and TuneCore. Downloads remain capped per plan, and every file is watermarked and fingerprinted. Automated mass exporting is explicitly targeted by these controls — the official route is the only sustainable one.

Are artists actually getting paid?

The labels are, from launch: Brody confirms that "a portion of the revenue" is shared with the partners, who then redistribute it. For individual artists, an opt-in program is in development: participation in remix features and compensation when their music is used. Believe/TuneCore artists can already opt in.

What are Suno's older models worth today?

Nothing: they were removed on the very day of v6's launch, according to CPO Jack Brody. Prompts and styles crafted on older versions don't produce the same results on v6, which was trained from scratch. Expect a relearning phase for your settings.

Will UMG and Sony sign next?

Nothing suggests so. Neither major is in v6's training set, neither has publicly reacted to their absence, and their lawsuits are ongoing. But Udio did sign with Universal after its own lawsuit: the precedent exists, and it makes a potential signing plausible.

✅ Conclusion

With v6, Suno turns a lawsuit into a licensed business model — and forces the entire generative AI industry to face the uncomfortable question: who pays creators, and how much. To keep up with what happens next (a possible deal with UMG and Sony, an opt-in program for artists, Udio's countermove), our selection of the best AI tools is updated as new launches roll out.