GitHub radar
YuE2: open music model that beats Suno on tests
A new open source 3.6B music generation model called YuE2 was released by m-a-p. You give it some text and a style description, and get back a full song including vocals and instrumental. It outperformed Suno v5 & v6 on SongBench.
YuE2 is an open source music generation model released by m-a-p research group. With a given text and style description, this model will generate a complete song including lyrics and musical instruments, with 48kHz 2 channels. This model has the capability of generating cover songs with various styles. Unlike other tools like Suno, this model also provides an editable song in ABC format, which means you could edit the melody and chords later. In SongBench Best of 8 benchmark, YuE2 gets a score of 6.9632, compared with Suno v5 (6.8721) and Suno v6 (6.5562). Generating a 3.6 minutes song would take around 71 seconds with RTX 4090. Under CC BY-NC 4.0 license, non-commercial use only.
Why a vibe-coder should care
If you are a content creator like Youtuber, podcaster, game developer, or social media user, and want to find some unique background music for your content, then YuE2 is the first open source tool that can compete with Suno on benchmarks. The feature of post-editing is really helpful. To use this model locally, you need a pretty strong GPU (24GB VRAM), so it might be more suitable for those who own such GPUs or can afford to pay for cloud GPUs.
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.
A GPU with 24 GB of VRAM (e.g. RTX 4090) and 24 GB of RAM are required — you can't run this on a regular laptop.
Open on Hugging Face▌ More finds