MiniMax's new-generation music model upgrades three things at once — how well it follows your intent, how the mix sounds, and how the vocals hold up. It's already selectable in SunoMV, so you can generate on it and get a lyric-synced music video out of the same session. Free to preview.
Music 3.0's three upgrades are not equally important to you. Intent-following saves regenerations and cleaner mixes are nice, but the rebuilt vocal engine is what decides whether a track survives contact with an audience — high-frequency vocal artefacts are the first thing listeners flag as 'AI'. If you are generating tracks to actually post, that is the upgrade worth switching for. If you only need a quick 30-second idea sketch, lighter and faster models in the list will serve you better.
MiniMax's own documentation puts the improvements in three places — and all three are exactly where AI-generated music usually gives itself away. Here's what that means when you're the one writing the prompt.
The upgraded semantic model reduces the usual AI-generated drift, per MiniMax. Write 'late night, restrained, just one piano' and you are less likely to get it buried in strings — which in practice means fewer regenerations.
MiniMax describes it as no more cramped or muddy mixes, with support for specific instruments and real techniques such as slides and legato. You can ask for 'electric guitar slide into the chorus' instead of vaguely writing 'rock'.
The high-frequency digital hiss is gone, and melody, pronunciation, breathing and layered harmonies become things you can steer. Vocals are the easiest tell in AI music, so this is the change that decides whether a track is postable.
Most places let you generate a song and stop there. SunoMV treats the song as raw material: pick MiniMax 3.0, write a line, and the lyrics get split into shots and turned into a music video you can post.
MiniMax 3.0 sits in SunoMV's model picker alongside the other music models — no waitlist, no API key, no setup. The previous 2.6 stays in the list if you prefer its character.
Write full lyrics if you have them, or a single sentence and let the AI expand it. Instrumental-only works too — skip the lyrics and describe the style.
Every line is auto-timed to the music and each section gets its own generated scene, so the finished video reads even with the sound off.
A song nobody sees is a song nobody hears. The output is shaped for where music actually gets discovered now.
One song, the right crop for TikTok, Reels and Shorts or a full-width YouTube upload — no re-rendering from scratch.
If you can pick a model and type a line, you can ship a music video. The syncing, pacing and rendering are automated.
Build and watch the finished lyric-synced video before paying anything. Exporting is part of the paid plans; previewing is free.
Nothing to install, no editor to learn, and no separate song-generation account.
Open the create page and select MiniMax 3.0 from the model list. Describe the style — name instruments and playing techniques, it understands them.
Paste finished lyrics, or write a single sentence and let the AI expand it. For an instrumental, skip the lyrics entirely.
The lyrics are timed to the beat and each section gets a generated scene. Preview the whole video free, then export vertical, square or widescreen.
No model is universally better — only better fitted. These are the cases where 3.0 is the one to reach for.
The vocal-engine upgrade removes the artefact listeners notice first, which matters far more for a release than for a scratch idea.
MiniMax has long been strong on Mandarin pronunciation and emotional delivery; the new vocal engine widens that lead.
If your brief is 'solo piano into the chorus' or 'electric guitar slide', 3.0 is the model that can act on that level of detail.
MiniMax Music 3.0 is MiniMax's new-generation text-to-music model, released August 2026. It generates a full song — vocals, lyrics and arrangement — from a style description and lyrics, and upgrades three things over 2.6: intent-following, mix quality, and a rebuilt vocal engine.
For tracks you plan to publish, yes — the rebuilt vocal engine removes the high-frequency artefacts listeners hear as 'AI'. Sound quality is subjective, so run the same lyrics through both and compare. In our measured run 3.0 also produced a longer, more fully structured arrangement from identical input.
Generate it inside SunoMV. Pick MiniMax 3.0 in the model list, write your lyrics, and SunoMV times each line to the beat and generates a scene per section, exporting a vertical, square or widescreen music video. Previewing the finished video is free.
Capability claims on this page come from MiniMax's own documentation. The timing and track-length figures are our own measured runs on integration day, not vendor numbers.
Music 3.0's three upgrades: reduced creative-intent drift, no more cramped or muddy mixes with support for specific instruments and techniques such as slides and legato, and a new vocal engine without high-frequency digital hiss.
MiniMax — Music Generation guide ↗music-3.0 is the current recommended music model in MiniMax's model line-up, alongside music-2.6 and music-cover.
MiniMax — Models overview ↗Tracks run up to five minutes; the API exposes instrumental generation and automatic lyric writing, but no duration parameter — length follows from the lyrics and their structure tags.
MiniMax — Music Generation API reference ↗FAQ'S
Ask us anything!
Pick it in the model list, write one line, and watch SunoMV turn it into a lyric-synced music video — free to preview, ready for TikTok, Reels, Shorts and YouTube.
/icon.png){kind=icon}
