MiniMax launches music generation model that creates full songs up to 5 minutes

AI company MiniMax announced the launch of MiniMax-Music3, a model capable of generating complete songs up to 5 minutes long based on lyrics and music descriptions. The model uses a two-component linguistic architecture: a global language model with 8 billion parameters that models long-range musical semantics and structure, and a local 600-million parameter model responsible for predicting detailed acoustic information. Output is generated as 32kHz stereo WAV files in 16-bit resolution. According to the company, the model maintains consistency in musical theme, rhythm, vocal identity, and arrangement progression throughout long pieces, covering typical structures such as intro, verse, pre-chorus, chorus, bridge, instrumental solo, and outro.