A 7 billion parameter instruction-tuned audio model from Moonshot AI, capable of understanding and generating audio content.