MiniMax Releases Open-Weights Music3 Model for Five-Minute Songs

Chinese artificial intelligence company MiniMax is extending generative AI from text and images into full-length music production. Its MiniMax-Music3 model uses an open-weights approach and accepts lyrics alongside structured style descriptions, lowering barriers for developers and businesses seeking to run music-generation systems on their own infrastructure. The release also puts questions around licensing, content governance and commercial use of AI-generated music back in focus.

MiniMax has formally released MiniMax-Music3, which can generate complete audio tracks of as long as five minutes in a single run rather than limiting output to short musical clips. The architecture combines two large language models with Flow-VAE technology to handle lyrical and musical structure. MiniMax released the model weights and inference code at the same time, allowing researchers, community developers and commercial users to deploy and adapt the system.