YuE2 (best-of-8) achieves 6.9632 on Song Bench, the highest among 15 evaluated settings on Wild Song Bench (192 prompts), outperforming Suno v5’s 6.8721. The model has approximately 3.59B parameters and 28 layers, supporting creation, covering, and editing of music through symbolic planning. MERT2-30s and MERT2-FS achieve state-of-the-art on 14 of 15 MARBLE metrics for music representation, leading in tagging, key, genre, and emotion recognition.

Sheet Sage2 attains SOTA on 10 of 13 benchmark metrics across beat, downbeat, key, chord, structure, and melody transcription, enabling audio-to-score conversion with vocal and instrumental melody voices. All components integrate symbolic scores, audio latents, and semantic tokens for end-to-end music generation and analysis.