Suno v6 lands: three music models trained on fully licensed catalogs
The clean-training-data angle that matters if you ship AI music — plus fine-tuning from your coding agent and fresh agent controls at Vercel and GitHub.
Copy markdown
Three models, one licensed dataset
Suno's v6, v6-wild and v6-mini were trained from scratch on catalogs licensed from Warner Music, BMG and Believe, and add prompt-based editing plus text, image and video reference inputs. If you put generated music in a product, a licensed-data foundation is cleaner legal footing than earlier models — though Suno's suits with Universal and Sony are still open.
Fine-tune from inside your coding agent
Fireworks shipped a training skill for Claude Code, Cursor, Codex and compatible agents: describe a fine-tuning goal in plain language and it plans the job, then waits for approval before running. A companion Cost Estimator previews spend up front — published per-token rates for Serverless/Managed, GPU-hour rates for Dedicated — so a training run stops being a blind bet.
Vercel locks down who wires up connectors
Vercel Connect now lets Pro and Enterprise teams restrict which members can manage connectors — a small governance switch that matters the moment your connectors touch production data.
Copilot starts metering your VS Code agents
GitHub added VS Code agent activity to Copilot usage metrics, so teams can see agent runs in the same dashboards they already use to track seats and spend instead of guessing at agent usage.