
Summary
- Suno released three variants on September 9: v6, v6-Wild, and v6-Mini. They were trained on recordings licensed from Warner Music Group, BMG, and Believe; Universal and Sony catalogs are absent.
- Chief product officer Jack Brody said v6 was trained entirely from scratch, and that a share of revenue goes to partners while rights holders decide how it is distributed. No percentages or amounts were disclosed.
- New features include section editing, image and video reference inputs, and stem separation from samples; remixing opens only when an artist opts in. Older models will be phased out.
Suno, the AI music service, released its new model v6 on September 9. The model was trained on recordings licensed from Warner Music Group, BMG, and Believe, and the company says it pays royalties to labels and publishers every time the model is used. Older models, it said, will be phased out gradually. This is the first time a company that makes music with AI has agreed to share money with rights holders and retrained a model on that basis.
Start with what changes for people making music. v6 comes in three variants: the base model v6 for paying Pro and Premier users, the experimental v6-Wild for the same paid tiers, and v6-Mini, available to all users. The company says v6 produces polished results regardless of genre, and introduced v6-Wild as less predictable but built for exploration. Of v6-Mini it said the model is better than competing models available for free.
The biggest change is the material it learned from. Chief product officer Jack Brody said v6 was "trained completely from scratch, from the ground up." He did not attribute the model's improvement to data alone, explaining that research and development and technical breakthroughs also played a part. Recordings from Universal Music Group and Sony Music are not included. Both companies are still in litigation with Suno.
The company also described how the money flows. Brody said "a portion of revenue is shared with partners, and how that capital is divided is up to the rights holders." It did not disclose percentages or amounts, nor how much reaches individual creators. Suno passes money to the rights holders, and the distribution after that is left to each company. Artists and labels participating through TuneCore are also inside this structure.
Features changed too. Specific sections of a song can be revised through prompts or lyric edits, and images and video, not just text, can be fed in as creative references. Instruments can be separated from a sample and used as material for a new song. A remix feature was also previewed, opening only when an artist chooses to opt in. Watermarking generated songs and tier-based download limits were measures already in place before this announcement.
Knowing where this company started changes the weight of this announcement. Suno has previously admitted to taking audio from YouTube videos while training earlier models, and the major labels have called this stream ripping and made it an issue. The models built from material gathered back then are the ones now being taken down. The company did not just swap out the model; it swapped out the material and rebuilt the model on top of it.
The litigation is not a single case. Suno settled with Warner Music Group in November 2025 and reached a partnership agreement with BMG in August 2026. The remaining cases keep piling up. The lawsuit the Recording Industry Association of America filed in June 2024 saw an amended complaint on August 25, and Suno filed its answer on September 1. It lost the German GEMA case on July 31; Denmark's Koda filed in Copenhagen in November 2025, and Canada's SOCAN filed on September 2. Round Hill Music filed on August 17, and singer Jason Isbell's class action on September 1. In July there was also a class action from users over a data leak.
The official announcement video METAL reviewed shows only screens introducing the three new variants, with no spoken narration. The company said it built the model together with artists and producers, but which artists took part and in what way is not in the announcement.
For people making content, two things are worth sorting out now. First, the models you have been using are going away. The same prompt will not produce the same sound as before, so if you have work in progress, it is better to capture the output first. Second, when the material changes, the sound changes. Because the model was retrained on licensed recordings, the texture of each genre may come out differently, and it is faster to generate a few new tracks and judge than to carry over your old prompts as they are.
To sum up: Suno released three variants of v6 trained on recordings from Warner Music Group, BMG, and Believe, is phasing out earlier models, and pays royalties to rights holders each time the model is used. Universal and Sony are still in litigation. Two things to watch from here: whether the revenue-share percentage is disclosed, and whether the two companies still suing join this structure.





Comments