Researchers have unveiled YuE2, a music generation system that aims to bridge two traditionally separate approaches: symbolic models, which make melody, harmony, rhythm and form explicit but stop short of a finished recording, and audio models, which produce complete songs while leaving composition implicit.
YuE2 unifies both through what the authors call symbolic planning. A single autoregressive–non-autoregressive Mixture-of-Transformers (MoT) first writes a readable score specifying melody and harmony, expands it into semantic music tokens, and finally realizes it as full-song audio with vocals and accompaniment.
In comparisons using the same checkpoint, experts preferred symbolic planning for overall quality and musicality, choosing it in 49.3% of overall preferences versus 34.6% without planning. Experts also favored the unified model over a separate language model and diffusion Transformer. On WildSongBench, YuE2 scored 6.73 on SongBench Global Avg, exceeding all evaluated public baselines; selecting from eight candidates (best-of-8) raised this to 6.96, the highest observed mean among all evaluated systems.
Expert listening tests further positioned YuE2 against proprietary song generators, favoring best-of-8 over Suno v4.5 and yielding nearly balanced preferences against Suno v5. To learn this generation process from recordings without aligned scores, the team introduced MERT2 and SheetSage2 to supply semantic and symbolic supervision. MERT2 sets a new state of the art in music representation learning, surpassing previous best results on 14 of 15 MARBLE metrics, while SheetSage2 leads 12 of 15 benchmark-metric pairs in the lead-sheet transcription comparison.
The same checkpoint follows score edits while largely preserving unedited musical content and generates zero-shot covers without cover-specific training. Its readable score also enables agentic music editing, with external language models translating user feedback into revisions of the composition. Model weights, demos, code and the technical report are publicly released.