Suno debuts v6 AI music model trained with record labels

Suno has officially released its v6 AI music model, marking the company's first release developed with direct support from the record industry. According to Suno, the model was built entirely from scratch using a brand-new dataset that excludes data from previous versions. Instead, this training set relies on licensed content from partners including Warner Music Group, BMG, and Believe, alongside user data, though it remains unclear if the training set is entirely free of unlicensed material.

The v6 release is split into three distinct models: v6, v6-wild, and v6-mini. The v6-mini model is available for free to all users, prioritizing fast and resource-light generation, though its outputs are simpler and contain more audible artifacts. Meanwhile, v6-wild is designed to introduce what Suno's Jack Brody described as "happy accidents and natural imperfections," though early testing shows it still struggles to deviate from rigid harmonic and rhythmic perfection.

While all three models demonstrate a vastly improved understanding of specific genres like hyperpop and krautrock, they still struggle to produce intentional discordance. Prompts requesting off-key instruments, monotone vocals, or out-of-tune pianos are generally ignored in favor of polished, mathematically perfect outputs. Additionally, vocal generations in v6 can sometimes exhibit harsher, more noticeable digital artifacts than those produced by the older v5 model.

For music creators, v6 introduces powerful workflow changes. Users can now edit specific parts of a song—such as a single guitar line or lyric—using plain-language chat commands without needing to regenerate the entire track. The model also allows users to mash together different elements from their existing Suno libraries. Furthermore, prompting is no longer limited to text, as v6 can generate music based on input images, video, or audio files. Suno has begun rolling out v6 today and plans to eventually retire its older models.