GitHub radar
YuE2: open music model that outscores Suno in tests
m-a-p released YuE2, an open 3.6B music generation model — text and style prompt in, full song with vocals and accompaniment out. It topped SongBench above Suno v5 and v6.
YuE2 is an open music generation model from the m-a-p research team. It takes lyrics text and a style description as input and outputs a complete song with vocals and instrumental accompaniment in 48kHz stereo. The model can generate cover versions of existing songs in a new genre. Unlike services such as Suno, YuE2 produces an editable score in ABC notation alongside the audio — letting you adjust melody and chords after generation. In SongBench tests (best-of-8), YuE2 scored 6.9632 against Suno v5's 6.8721 and Suno v6's 6.5562. A 3.6-minute song takes roughly 71 seconds on an RTX 4090. License is CC BY-NC 4.0 — non-commercial use only.
Why a vibe-coder should care
If you create content that needs original music — YouTube, podcasts, games, social — YuE2 is the first open alternative to Suno that reaches comparable quality in benchmarks. The editable score is a practical advantage: you can adjust the melody after generation, which Suno doesn't offer. Running it locally requires a high-end GPU (24GB VRAM), so it's mainly relevant for those with the right hardware or access to cloud GPU rentals.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Check how to try the YuE2 music model from m-a-p: https://huggingface.co/m-a-p/YuE2-3B — is there a browser demo space, how to run it locally with a GPU, and what to write to generate a first song.
Нужна видеокарта от 24 ГБ видеопамяти (например, RTX 4090) и 24 ГБ оперативной памяти — на обычном ноутбуке не запустить.
Open on Hugging Face▌ More finds