Qwen Enters Music Generation With Qwen-Music

Alibaba's Qwen team has released Qwen-Music, its first model aimed at generating complete songs. According to the technical report, the model produces high-fidelity audio that includes sung vocals, taking text descriptions, lyrics, and musical prompts as inputs.

The pitch is straightforward: rather than generating instrumental beds or short loops, Qwen-Music targets full songs where the words a user supplies are actually sung. That combination of lyric-conditioned vocals and instrumental backing places it alongside a small but growing group of systems attempting end-to-end music creation.

Why it matters

Qwen has become one of the most prolific open-model families, spanning text, vision, and audio. Extending the brand into music signals that Alibaba sees generative audio as part of its broader multimodal strategy, and it gives developers another option outside the handful of well-known music generators.

• Inputs span text, lyrics, and musical prompts
• Output is high-fidelity song audio with vocals
• Released under Qwen's own license as a version 1.0

As an initial release, the details that will matter most — training data, evaluation against rival systems, and practical licensing limits — are worth scrutinizing as the model reaches wider use. For now, Qwen-Music establishes a foothold in a category the team has not previously entered.