On March 26, Kunlun Wanwei officially launched two new music generation models: Mureka V6 and Mureka O1. Notably, Mureka O1 is recognized as the world's first large-scale music reasoning model to incorporate Chain-of-Thought (CoT) technology, marking a significant advancement in the field of AI-generated music.
Mureka V6 serves as the foundational model, supporting both instrumental music creation and lyric writing in ten languages including English, Chinese, Japanese, Korean, and French. Its proprietary In-Context Learning (ICL) technology has notably improved sound field performance, vocal texture, and mixing design effects. Users can generate multi-style music through simple text input, covering genres such as jazz, electronic, pop, and rock, while supporting diverse emotional expressions.
Mureka O1 builds upon the V6 model with optimizations, introducing CoT technology into music generation for the first time. This technology enhances audio coherence and instrument arrangement accuracy by pre-generating overall musical structures. Compared to traditional step-by-step generation methods, O1 shows significant improvements in the alignment of lyrics and melody, vocal accuracy, and artistic expression. According to public data, Mureka O1 outperforms Suno V4 in multiple objective tests, particularly excelling in mixing quality, vocal texture, and background music.
In addition, the Mureka platform offers two distinctive features: song referencing and voice cloning. The song reference function allows users to upload audio or links as prompts to generate music in a similar style directly. Voice cloning enables users to select from official singer voices or upload their own recordings to create highly realistic personalized works. These features lower the technical barriers of professional music production, empowering ordinary users to participate in music creation easily.
For developers, Mureka provides API services and model fine-tuning capabilities. The API supports applications in music generation and speech synthesis, suitable for content creation, game soundtracks, and short video production. Users can customize exclusive music styles by uploading private data for model fine-tuning. Regarding data privacy, the platform guarantees strict protection of uploaded user data, ensuring its use only for personal model training.
The Mureka research team has disclosed the core technical principle of the O1 model—the MusiCoT algorithm framework—aiming to promote industry-wide progress. Based on the CLAP model, MusiCoT expands without requiring manual annotation, significantly improving the explainability and quality of generated music. This technological innovation represents a major breakthrough in high-fidelity AI music generation.
