Suno has confirmed that its V6 model family was trained from scratch on a new dataset combining content licensed through partnerships with Warner Music Group, BMG and Believe with what the company describes only as “user data.” The disclosure confirms that user-derived information contributed to the training mix, but Suno has not publicly clarified whether that category includes community-generated songs, uploaded audio, prompts, preference signals or other platform activity. Suno maintains that V6 was not trained on the same dataset used for its earlier models.

That distinction has not resolved Suno’s copyright dispute with Universal Music Group and Sony Music. In a September 18 complaint, the labels allege that V6 still derives value from outputs, user interactions and capabilities transferred from earlier Suno models that they claim were trained on unlicensed recordings. Calling V6 “the fruit of the same poisoned tree,” the labels argue that rebuilding the model on a new dataset cannot erase infringement allegedly embedded in the earlier system. Those claims remain allegations rather than court findings; Suno maintains that V6 represents a new training process and has separately argued that its earlier training practices qualify as fair use.