Suno v6 retrained on licensed music. Check your lineage
Suno rebuilt v6 from scratch on licensed Warner, BMG and Believe catalogs and retired its old models the same day. Training-data provenance is now a product feature.
Suno launched v6 on September 9 and did something no major generative vendor had done at this scale: it threw out the training data. The new models were built from scratch on licensed music from Warner Music Group, BMG, and Believe — explicitly not on the corpus behind every previous Suno model — and the old generation was retired the same day. If you use generative AI in anything you sell, this is the moment training-data provenance stopped being a philosophical argument and became a shipped product decision.
What actually happened
Three models: v6 and v6-wild for Pro and Premier subscribers, v6-mini for everyone. CPO Jack Brody says v6 "was trained entirely from scratch, from the ground up," and Music Business Worldwide reports the training set covers WMG, BMG, and Believe — including TuneCore artists — while excluding Universal Music Group and Sony Music catalogs. Revenue sharing starts at launch, with the partners responsible for paying rightsholders downstream.
The legal backdrop explains the rebuild. Suno settled with Warner in November 2025 and reached a deal with BMG in August 2026. Sony Music, Universal, and artist Jason Isbell are still litigating, and a user class action over the platform's data breach is open. Retraining doesn't retroactively clean the old models — it separates the going-forward product from the claims.
The release also adds prompt-based editing of existing songs, text/image/video reference inputs, and stem separation for remixing. Watermarking and tier-based download caps, which Suno announced in August, carry forward.
Why it matters for your business
Two things happened here at once, and the smaller one is the trap.
The retirement is the risk. Suno turned off the previous generation on launch day. Any workflow that depended on a specific model's sound just changed underneath it, with no migration period. This is the third time this year we've written some version of that sentence about a different vendor. If AI output is an input to something you deliver, treat the model version like a dependency: pin it where you can, keep the source prompt and seed, and archive final renders as files rather than as "we can regenerate it."
The licensing is the opportunity. A vendor that can name its training sources can offer terms about output. That's the axis commercial buyers will start shopping on — not benchmark scores. When you evaluate any generative tool for work that leaves your building, ask three questions and get the answers in writing: What was it trained on? What rights do I have to the output? Does that survive my subscription ending?
Most vendors currently answer the first with silence. Suno just made silence a competitive disadvantage, and that's good for everyone buying.
Key takeaways
- Suno released v6, v6-wild, and v6-mini on September 9, 2026, trained from scratch on licensed WMG, BMG, and Believe catalogs
- UMG and Sony catalogs are excluded; both labels, plus artist Jason Isbell, are still litigating against Suno
- Suno retired its previous model generation the same day — no migration window for existing workflows
- Revenue sharing with label partners begins at launch, with partners distributing to rightsholders
- Treat model versions as pinned dependencies and archive final assets as files, not as regenerable prompts
If AI output ships to your customers, you need to know where it came from and what you own. We build content pipelines with pinned models, archived assets, and a paper trail per deliverable. See how we handle AI in production work, or ask us to audit what your current stack actually licenses you.
Sources: Music Business Worldwide, TechCrunch.
- #ai-content
- #licensing
- #provenance
- #vendor-risk
- #generative-ai
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Mercury 2.5 hits 1,107 tokens/sec: price the fast path
Inception's Mercury 2.5 diffusion LLM runs 1,107 tokens per second at $0.20/$0.75 per million. Latency is now a routing decision, not a model you are stuck with.
Read itClay's $115M Series D at $7.1B: outbound isn't the moat
Clay raised $115M at a $7.1B valuation with 17,000+ teams on the platform. The GTM automation layer is commoditizing — here's what still belongs to you.
Read it