Suno has released v6, a new generation of its model designed to generate music from text prompts. What sets it apart from previous versions is not just the quality of the generated tracks: it is the company's first model developed with the support of the recording industry and trained on a new dataset that includes licensed content from Warner Music Group, BMG, and Believe.
For an industry that has long debated copyright, remuneration, and the use of creative works in training generative systems, this is a significant step. Suno claims to have built v6 from scratch, without reusing the dataset employed for prior models. Jack Brody, an executive at the company, told The Verge that the new system relies on different data. However, that phrasing does not fully clarify the origin of all the materials used.
Alongside catalogs licensed by record label partners, Suno also mentions “user data.” Not enough details have been provided to determine whether the v6 dataset consists exclusively of content with clearly cleared rights or if other sources are involved. It is an important distinction: bringing major labels' licenses on board marks a shift in relations with the music industry, but it does not automatically translate into complete transparency across the entire training process.
Three models, three trade-offs
The new release does not arrive as a single engine. Suno is splitting the offering into three variants: v6, v6-wild, and v6-mini. This division aims to address different needs, ranging from faster and more accessible generation to the pursuit of less predictable results.
- v6-mini is available for free and focuses on speed and lighter resource usage;
- v6 is the flagship model of the new family;
- v6-wild is designed to introduce greater unpredictability into the results, including those accidental variations that can make a performance feel less uniform.
The mini model pays for this immediacy with simpler tracks and a higher presence of artifacts traceable to automated generation. It is an expected trade-off for a tier accessible to everyone, but it also makes clear that the v6 name does not denote an identical improvement for every user and subscription plan.
More interesting, at least on paper, is v6-wild. Suno presents it as a variant capable of leaving room for happy accidents and more natural imperfections. In tests reported by the US publication, however, the difference compared to standard v6 proved difficult to discern. The risk is that the promised higher unpredictability remains more of a design goal than an easily perceptible quality in the generated tracks.
Genres improve, human error still doesn't
Where v6 does seem to take a step forward is in its grasp of genre conventions. Compared to previous versions, the system recognizes the surface and stylistic cues associated with prompts like hyperpop or krautrock with greater accuracy. For those using these tools for sketches, demos, temp tracks, or idea exploration, this is a tangible improvement: a genre description is more likely to yield arrangements, sounds, and an attitude consistent with what was requested.
Yet fidelity to musical tags does not equate to truly simulating a musician's behavior. According to tests cited by The Verge, v6 continues to steer performances toward an almost stubborn regularity. Explicit prompts for off-key notes, mild dissonances, out-of-tune pianos, or playing without regard for rhythm and key failed to produce the desired effect. Instructions such as a monotone vocal or an absence of drums can also be ignored by the model.
It is a limitation less marginal than it might seem. Much of music does not thrive solely on tidy harmonies and perfectly aligned tempos: hesitations, irregular accents, slightly rushed or dragged cues, and timbral flaws are expressive tools. A system that readily absorbs the outward traits of a genre, yet tends to normalize the performance, can produce material that sounds convincing on first listen without ever reaching the interpretive complexity of a human recording.
Paradoxically, Suno v6 still retains some of the imperfections inherent to AI-generated music. Vocals, in particular, can exhibit harsher-sounding artifacts compared to v5. On the one hand, then, the model struggles to produce the intentional flaws requested by the user; on the other, it does not completely eliminate the unwanted anomalies that reveal a track's synthetic origins.
The weight of licensing beyond the product launch
The most significant aspect of v6 may therefore lie less in its new features than in how it was built. Warner Music Group, BMG, and Believe are key partners because they bring players representing a substantial share of the recording industry into Suno's trajectory. For companies developing music models, being able to point to licensed training content is valuable both commercially and in terms of legitimacy among artists, publishers, labels, and platforms.
The licensing model, however, raises operational questions that this presentation leaves unanswered. What content was included, under what terms, how authors' contributions are handled, and what control exists over the use of user data are all factors directly affecting trust in the platform. The mention of a new dataset marks a break with the past, but on its own does not make it possible to trace the full pipeline of the material used.
For Suno, the launch is therefore a two-fold test. It will have to prove that v6 offers a meaningful leap forward for music creators, without merely replicating genre tropes with greater accuracy; at the same time, it must lend credibility to the idea of a music AI developed in concert with the recording industry. Early assessments suggest a system that is more effective at following stylistic coordinates, but still difficult to control when users seek the messiness, fragility, and imperfection that give a performance its character.
Meanwhile, the availability of v6-mini extends access to the new model family to free users as well. It remains to be seen whether Suno will accompany this rollout with more precise information about data composition and whether the wild variant will manage, through subsequent updates, to truly set itself apart from the main model. For now, v6 primarily marks the start of a phase in which output quality and data provenance become two inseparable parts of the same product.
