MiniMax opens a five-minute music model

MiniMax has released Music 3, an open music-generation model that creates complete songs up to five minutes long with vocals, evolving arrangements, and 32 kHz, 16-bit stereo output. The official model card describes an 8B global language model paired with a 0.6B local model and says the stack can fit under 24GB of VRAM at full precision, with CPU offloading lowering the GPU requirement to about 8GB at slower speed. Creative-tool builders can now test long-form local music generation without an API dependency, but the card’s caveats—CUDA-only, non-streaming inference, and no strict guarantee that lyrics, tempo, or structure will be followed—make it a prototyping surface rather than a finished production primitive, extending the archive’s open-model economics thread.