AI Expertise
Open Models / YuE2-3B
Y

YuE2-3B

Visit website

Description

YuE2-3B is an open music model from the M-A-P community that creates a full song with vocals from a style description and lyrics. Per the model card, the output is 48 kHz stereo; the model supports Chinese and English, can make a cover of an existing song and exports a plan of the song (in ABC notation) that you can edit and generate again. Released on 2026-09-09; it continues the original YuE of January 2025. The weights are public on Hugging Face without any acceptance step. The weights are licensed under CC BY-NC 4.0 (see the license file in the model repository): per the license text, commercial use is not allowed and attribution is required. The code in the authors' GitHub repository is licensed separately under Apache 2.0 — that does not apply to the weights. The card recommends a 24 GB GPU; the authors report a peak of about 11 GB on an RTX 4090. The authors also have a demo page with examples to listen to and a paper on arXiv. We did not find an official Space from the authors. The card gives the authors' own benchmark results; we do not repeat them as fact. Date of check: 2026-10-18.

Pros and cons

Pros

  • Full song with vocals from a style description and lyrics
  • 48 kHz stereo, per the model card
  • Cover mode and an editable song plan for regeneration
  • Weights open without an acceptance step
  • Demo page with examples to listen to

Cons

  • Per the license text, commercial use of the weights is not allowed (CC BY-NC 4.0)
  • Only Chinese and English are stated
  • A 24 GB GPU is recommended
  • No official Space from the authors found

Key facts

License
CC BY-NC 4.0
Commercial use
Not allowed
Hardware requirements
Per the model card: a 24 GB GPU and 24 GB of RAM are recommended; the authors report a peak of about 11 GB on an RTX 4090
Developer
M-A-P (Multimodal Art Projection)
Interface languages
English
Checked on
Oct 18, 2026

Editorial score

How strongly the AI Expertise editors recommend this open model for each of its tasks right now, on a 0–10 scale. The score combines output quality, capabilities, ease of launch, license openness and freshness.

Music

7.3 / 10
Output quality 8.5
Low High
Features 9.0
Basic Rich
Ease of launch 7.5
Hard Easy
License openness 2.0
Restricted Open
Freshness 10.0
Outdated Fresh