MiniMax introduces Music 3.0 for full-song generation
The model combines structured musical descriptions with a redesigned generation and audio-rendering pipeline.
THE STORY IN THREE POINTS
- MiniMax announced Music 3.0 on August 13.
- The company describes generation from a concept and optional lyrics.
- Its quality and consistency claims require independent listening tests.
What was announced
MiniMax introduced Music 3.0 on August 13 as an open-weights music-generation model. The company says it can compose and render a complete song from a creative concept and optional lyrics, with outputs up to five minutes. The release emphasizes maintaining musical intent, arrangement, instrumental detail, and vocal delivery across a song.
The details that matter
Its technical account describes structured captions, a global-local language-model architecture, and an audio-synthesis stack. The company’s examples illustrate the intended behavior; they do not establish consistent performance for every prompt. August 13 is the source publication date, rather than a new release attributed to today.
Editorial context
A meaningful listening test should include complete songs rather than only the strongest short extracts. Reviewers could examine whether an instrument persists when requested, whether sections develop coherently, and how often a creator must regenerate before obtaining usable material. Our editorial interest is the distance between a musical brief and the full delivered track. Access to model weights can support experimentation, while the rights and permitted uses attached to any actual deployment still need to be read in the applicable release terms and agreements.
Sources: MiniMax: Music 3.0: Next-Generation Open-Weights, Production-Ready & Versatile Music Model ↗
