Music generation AI company Suno announced on September 9, 2026, a new generation of models called "v6," developed in partnership with Warner Music Group (WMG), BMG, and Believe. The rollout includes three distinct models for different use cases, supporting natural-language partial editing and multi-format inputs. As part of the transition, Suno plans to retire its previous models and move its entire service to the v6 generation.
On the surface, this looks like a generational update improving audio quality and usability. But the real turning point is that Suno—which admitted in 2024 to training on music scraped from the open internet—is now migrating its entire user base to new models built under licensing agreements with rights holders. The company has begun rebuilding not just the models themselves, but the surrounding distribution and revenue infrastructure, including provenance tracking, download limits, and future artist-centric products.
Three Models Split Precision Control From Creative Randomness
The standard v6 model, available to Pro and Premier subscribers, is described by Suno as reliably delivering intended results. The v6-wild model, available under the same two plans, deliberately preserves unpredictability and variation to expand creative possibilities. All users, including free-tier subscribers, get access to v6-mini, a faster and lighter-weight version.
The core of the new functionality shifts the workflow from one-shot song generation to interactive editing of in-progress material. For example, users can transform just the chorus of an existing song into a choral arrangement while leaving other sections untouched. The system is also designed to let users extract vocals and drums from multiple original tracks and combine them into one, or sample a guitar part from a specific moment to build a new beat around it.
Inputs aren't limited to text and audio, either. The update supports generating songs from multiple formats including images and video, and even allows swapping out a single word in lyrics. This update pushes Suno closer to being a production environment where users move back and forth between materials to finish a song, rather than simply a generative model.
However, Suno's claims that the new models are "faster," "higher quality," and "more expressive" come without comparison baselines, testing conditions, metrics, or measured results. The claim that v6-mini is the best free music generation model is also the company's own self-assessment, not an independent benchmark. While the expanded feature set is verifiable, the actual scope of performance improvement remains unverified.
Warner, BMG, and Believe Play Different Roles
While the three companies are collectively described as "industry partners," their documented roles differ substantially. WMG signed an agreement with Suno in November 2025 to jointly develop next-generation licensed models and retire the older ones. That agreement also settled prior litigation between the two companies. WMG artists and songwriters can choose whether to allow their names, images, likenesses, voices, and songs to be used in AI-generated tracks.
BMG's agreement, signed in August 2026, covers both the recorded music and music publishing repertoire the company manages. It protects the rights of participating artists and writers, provides compensation for use of their music, and also resolves past use of BMG's catalog. This carries two distinct implications: future use of training data, and settlement of past usage.
Believe and its subsidiary TuneCore connect independent artists and labels to the new models. Music from participating artists will be included in Suno's new models, with the companies jointly developing licensed products. Additionally, songs created using the music-industry partnership models will now be distributable through Believe and TuneCore. This marks a clear reversal from April 2026, when Believe had refused to distribute songs made with Suno's then-current models.
In short, WMG's role centers on v6's initial training and settlement with a major label; BMG's covers both recording and publishing rights clearance; and Believe's focuses on independent catalog access and distribution channels. The fact that all three companies signed agreements doesn't mean their respective catalogs entered v6's training at the same time or to the same extent.
Even With Licensing, the Full Training Data Picture Remains Unclear
Suno's Chief Product Officer, Jack Brody, told Axios that v6 was built from scratch, trained on a combination of Warner-licensed music and Suno user data. He did not say that the weights or training data from previous models were carried over directly, but said technical know-how and insights into user preferences were retained. He also said that prompts attempting to directly mimic known artists or songs will still be blocked in v6.
Based on publicly available information, the only confirmed training inputs for v6 are Warner-licensed music and Suno user data. The total size and composition of the training corpus, the completion timeline for integrating BMG and Believe catalogs, and the specific metrics and figures behind performance claims have not been disclosed. While BMG's and Believe's announcements indicate an intent to add their participating catalogs to the new model, they don't specify what percentage of that catalog was actually incorporated by the time v6 launched, or how many tracks that represents.
This gap matters for measuring how different v6 actually is from the earlier model era. In 2024, Suno admitted to training on medium- and high-quality music found across the open internet, acknowledging that this included copyrighted material. At the time, the company argued that this training did not constitute infringement, comparing it to the way humans learn by listening to music.
With v6, at least the Warner-sourced material is now governed by an agreement made in advance. But the label "licensed model" alone doesn't guarantee that every piece of the training corpus was authorized. Which data was used, under what rights, and at what point it entered training—these questions form a chain that must be disclosed before anyone can verify whether Suno has truly broken from the issues surrounding its previous models.
Provenance Tagging and Download Limits Tighten Control on Output
Separately from clearing rights on training data, Suno has introduced mechanisms to attach provenance information to generated songs and limit how many times they can be exported externally. Generated tracks now carry Content Credentials (C2PA-compliant content credentials). Even if a file is taken outside Suno's platform, compatible distributors and verifiers can machine-check its origin. Suno also says it will check user-uploaded audio and lyrics for similarity to existing works.
Download caps also went into effect on September 3. For free users who registered before that date, the lifetime trial allowance is capped at 7 songs total; Pro subscribers get 20 songs per month; Premier subscribers get 60 songs per month. Free users who registered after September 3 may receive trial allowances issued periodically. There is no cap for Premier subscribers exporting from Suno Studio. The free tier is limited to personal, non-commercial use, while songs downloaded under paid plans come with commercial usage rights. Users can purchase additional downloads beyond the cap.
This limit applies not to the number of generations, but to how many times files are exported outside the platform. Users can still play songs on Suno and continue sharing links indefinitely. Suno describes the limit as a measure to make it harder for bad actors to bulk-export large volumes of songs. However, Suno Studio has no cap, and there's no measured data yet demonstrating the limit's deterrent effect. Whether revenue from additional download purchases will go toward licensing fees or rights-holder payouts has not been disclosed.
That said, provenance information is a tracking mechanism—including safeguards against deletion or tampering—not a guarantee that infringement will be fully prevented. Similarity checks can also produce false positives and false negatives, though Suno hasn't published accuracy figures. Additionally, even if Suno grants commercial usage rights, whether a generated song receives copyright protection in a given country, or whether it infringes on third-party rights, is determined separately.
What to Watch Next: Catalog Integration Volume and Revenue-Sharing Formulas
As a next step, Suno is developing new artist-centric products. The vision involves artists opting in individually, and compensating them when fans engage with new experiences involving their voice or songs. This isn't a feature already shipped as part of standard v6—it's a future plan built on top of the current model.
Brody told Axios that revenue sharing is included in the agreements with rights holders, but did not disclose the specific rate or which types of usage would be subject to sharing. Whether the denominator is generation counts, downloads, subscription fees, or future artist-specific products will significantly affect how much value rights holders actually receive.
Whether v6 can demonstrate a genuine break from the earlier model era won't be settled by naming alone. First, there's the question of when—and at what scale, in terms of track counts—BMG's and Believe/TuneCore's participating catalogs actually enter training. Second, how much Suno discloses about its revenue-sharing formulas with rights holders and its handling of user data. Third, whether v6, once the older models are retired, can be measured under independent conditions to demonstrate concrete improvements in audio quality and prompt-following accuracy. Only when all three of these come together will the agreements with rights holders move from promises in a press release to a system that users and musicians can actually verify.