MiniMax released MiniMax-Music3, an open-weights text-to-music model. The model takes two separate inputs: lyrics carrying section tags, and a detailed music description. It returns a complete song of up to five minutes in a single generation, as 32 kHz, 16-bit stereo WAV. The architecture pairs a Hybrid-LM, an 8B Global LLM with a 0.6B Local LLM, with a continuous synthesis stack built on flow matching and a Flow-VAE. Weights, inference code and three documented serving paths shipped the same day.
Is it deployable?
Yes, MiniMax published usable weights, inference code and three documented serving paths on day one, so this is deployable now rather than a research preview.
- Company level: Solo creators, indie studios and mid-market teams can ship on it directly. The MiniMax-Music3 Community License permits commercial use, but it requires you to display ‘MiniMax-Music3’ prominently in the product UI, and any organization whose aggregate yearly revenue from those products exceeds US$ 20 million must obtain separate prior written authorization from MiniMax. Anyone hosting third-party generation must also implement and maintain safeguards against infringing outputs.
- Industries: Game development, advertising and brand agencies, short-form video and creator tools, e-learning, podcasting, fitness and wellness apps, retail in-store audio, and music-tech SaaS.
- Applications: Background scoring for UGC video, adaptive game and level music, localized ad beds and sonic branding, scratch and demo tracks for songwriters, mood-conditioned playlist generation, and offline batch generation where per-song API cost is the constraint.
The Architecture
MiniMax-Music3 combines a hierarchical autoregressive stack with a continuous synthesis path.
The training tokenizer uses eight layers of residual vector quantization (RVQ). The first, semantic codebook has 16,384 entries and carries core musical semantics and structure. The remaining seven acoustic codebooks have 1,024 entries each and encode residual detail. Training optimizes the semantic layer first, then all eight jointly.
The Hybrid-LM splits the modeling problem. An 8B Global LLM predicts the first RVQ codebook frame by frame and holds long-range structure; a 0.6B Local LLM predicts the remaining codebooks within each frame. The model card and license state the Global LLM is initialized from Qwen3-8B; the MiniMax Research post says Qwen3.5-8B, so treat the exact base checkpoint as unsettled.
The synthesis stage is the more interesting design choice. Rather than decoding from discrete RVQ tokens, MiniMax fuses the final hidden states of both LLMs and conditions a 2.4B flow-matching module on them, which maps into a latent space decoded by a 123M Flow-VAE inherited from MiniMax Speech. At inference the discrete tokenizer decoder is not loaded at all.
Two-input control
Lyrics carry the words and section tags on their own lines: [Intro], [Verse], [Pre-Chorus], [Chorus], [Post-Chorus], [Bridge], [Instrumental], [Solo], [Outro]. A separate Structured Caption carries Global Metadata, Vocal Details and Arrangement. MiniMax also ships a music-caption-rewriter agent skill that expands a short description into that three-part format offline.
Interactive explainer
‘;
if(i
