YuE2 · Frontier Music with Symbolic Planning

From score to song

Listen to a song, then explore the melody, rhythm, and chords in its symbolic plan.

Selected score

Loading selected song…

The generated song

The symbolic score

Original score recording

View original score pages

All selected scores

Cover & Editing

A familiar song can take a different shape. Listen to changes in melody, lyrics, tempo, and arrangement.

Agentic music editing

A song takes shape through a conversation. Loading the editing story…

Genre Explorer

The listening selection, gathered across genres and languages.

Model & Results

YuE2 (best-of-8) reaches 6.9632 on SongBench, the highest observed mean among 17 evaluated settings on WildSongBench (192 prompts). Suno v5 scores 6.8721 in the same comparison. Suno v6 and Suno v6 Wild score 6.5562 and 6.4195, respectively.

WildSongBench comparison: 17 system settings, including Suno v6 and v6 Wild, compared on song quality and text alignment. YuE2 and YuE2 best-of-8 are competitive with the evaluated proprietary systems. Bubble area represents AudioBox production quality. — Song quality and text alignment on WildSongBench. Bubble area shows AudioBox PQ; black outlines mark Pareto optima on the two plotted axes. Bo8 = best-of-8. How to read the indices.

YuE2 architecture: an AR–NAR Mixture-of-Transformers turns an editable symbolic score into semantic tokens, acoustic latents, and full-song audio. The same generator supports creation, covering, and editing. — An editable score becomes semantic music tokens, acoustic latents, and full-song audio. The model has approximately 3.58B parameters and 28 layers, and supports creation, covering, and editing. The AR and NAR experts share an attention computation while using separate normalization, projections, and MLPs.

Explore benchmark scores WildSongBench · 17 settings · 9 metrics

WildSongBench192 prompts

Compare by

WildSongBench · SongBench
Rank	System / setting	Prompts	SongBench ↑

Updated September 12, 2026 · rankings vary by metric.

Download results

How to read these results

WildSongBench. 192 prompts, 17 system settings, and the paper’s nine automatic metrics. SongBench averages all seven dimensions; SongEval averages all five. Musicality is also reported separately for each evaluator. Both YuE2 settings use melody-and-chord symbolic planning.

Selection. YuE2, the public baselines, and Suno v6 and v6 Wild use two candidates and four ASR repeats per candidate, followed by lower-PER selection. Earlier proprietary outputs retain their delivered-candidate protocols. YuE2 best-of-8 selects from eight generations using SongBench Musicality, then Q3O Control, then PER. The comparison therefore retains different selection budgets.

MERT2 architecture: offline target synthesis combines MuQ and Qwen2-Audio features into four code streams. A ConvNeXt frontend and 24-layer Conformer learn by masked prediction, then branch into full-song representations for SheetSage2 and a causal tokenizer curriculum for YuE2. — MERT2 provides the music representations behind both analysis and generation. A ConvNeXt frontend and 24-layer Conformer learn to predict masked codes built from complementary MuQ and Qwen2-Audio features. Full-song adaptation supplies SheetSage2 with musical context; a separate causal branch becomes YuE2’s 25-Hz semantic tokenizer.

State of the art on MARBLE

MERT2-30s achieves SOTA on 14 of 15 MARBLE metrics against the listed external baselines. MERT2-FS (full-song) exceeds those baselines on 13 of 15 metrics.

Leading metricsMERT2-30s · versus external baselines: 14 / 15
Leading metricsMERT2-FS · versus external baselines: 13 / 15
Key refined accuracy · GiantStepsMERT2-FS · score × 100: 67.05

Explore MERT2 benchmark scores MARBLE · 15 metrics · 2 encoders

Scores × 100 · higher is better · bold marks the highest displayed score.
Benchmark / metric	Best published baseline	MERT2-30s	MERT2-FS
MTT · TaggingROC-AUC	91.70PupuJEPA-Large	91.91	91.74
MTT · TaggingAverage precision	40.80PupuJEPA-Large	41.29	41.20
GiantSteps · KeyRefined accuracy	66.10PupuJEPA-Large	66.97	67.05
GTZAN · GenreAccuracy	86.90PupuJEPA-Large	91.72	90.69
GTZAN · BeatF1	91.00PupuJEPA-Large	90.59	90.57
EmoMusic · ValenceR²	62.50PupuJEPA-Large	63.23	63.52
EmoMusic · ArousalR²	78.50PupuJEPA-Huge	80.01	78.14
MTG-Jamendo · InstrumentROC-AUC	78.40PupuJEPA-Large	80.27	80.27
MTG-Jamendo · InstrumentAverage precision	21.20PupuJEPA-Large	22.89	23.51
MTG-Jamendo · Mood / themeROC-AUC	76.20PupuJEPA-Large	79.44	78.74
MTG-Jamendo · Mood / themeAverage precision	15.50Dasheng-1.2B	16.68	15.74
MTG-Jamendo · GenreROC-AUC	86.30AudioMAE++	88.01	87.98
MTG-Jamendo · GenreAverage precision	20.10PupuJEPA-Large / PupuJEPA-Huge	21.22	20.66
MTG-Jamendo · Top 50ROC-AUC	83.10AudioMAE++ / PupuJEPA-Huge	84.18	84.13
MTG-Jamendo · Top 50Average precision	31.10AudioMAE++	32.17	31.62

Counts compare each MERT2 encoder separately against the strongest of the nine external baselines for each metric; the two MERT2 encoders are not counted as baselines for each other. Both encoders have 632M parameters. MERT2-30s uses 30-second training clips; MERT2-FS continues pretraining on complete recordings lasting 30–360 seconds. These benchmarks evaluate bidirectional representations, not the causal tokenizer. MERT2 reports the best observed results across representations; each ROC-AUC / AP pair uses the same representation.

Baseline scores follow Tables III and V of the published comparison. Naming note: the authors renamed PupuJEPA to PupuM2D after the paper’s initial release; we retain PupuJEPA.

Published baseline comparison ↗

Download all 11 models

SheetSage2 architecture: a full-song MERT2-FS encoder with trainable adapters feeds a six-layer autoregressive decoder. Task prompts select beat, section, key, chord, and melody events, which share a timeline and become ABC notation and a lead sheet. — SheetSage2 turns a recording into an editable lead sheet. A full-song MERT2-FS encoder, adapted with LoRA, feeds a six-layer autoregressive decoder. Task prompts select beats, sections, keys, chords, and melodies; a shared event timeline becomes ABC notation with vocal and instrumental melody voices. These scores supply symbolic training targets for YuE2.

Six transcription tasks, one model

SheetSage2 achieves SOTA on 12 of 15 benchmark metrics in the displayed comparison, with one autoregressive model (SheetSage2-AR) for beat, downbeat, key, chord, structure, and melody transcription.

SOTA metricsOne model · six transcription tasks: 12 / 15
Vocal melody · RWC-PopPitch-class note F1 · score × 100: 82.51
Chord recognition · osu2017Maj/min · score × 100: 90.08

Explore SheetSage2 benchmark scores 6 tasks · 15 metrics

Scores × 100 · higher is better · bold marks the highest displayed score.
Benchmark / task	Metric	Best comparison	SheetSage2-AR
GTZANBeat	F1	89.01Beat This!	86.27
osu2017Beat	F1	91.80Madmom	93.01
GTZANDownbeat	F1	78.28Beat This!	80.45
osu2017Downbeat	F1	85.90Beat This!	92.90
GiantStepsKey	Weighted score	74.62Madmom	77.73
GTZANKey	Weighted score	74.43S-KEY	75.77
osu2017Chord	Maj/min	86.55ChordFormer	90.08
Chords1217Chord	Maj/min	83.94ChordFormer	83.81
JAAHChord	Maj/min	59.45ChordFormer	64.50
HarmonixSetStructure	Accuracy	80.03SongFormer	80.51
HarmonixSetStructure	Boundary F1 · 0.5 s	70.63SongFormer	67.96
HarmonixSetStructure	Boundary F1 · 3 s	79.50SongFormer	82.86
RWC-PopMelody	Vocal pitch-class F1	62.71SheetSage1	82.51
RWC-PopMelody	Full pitch-class F1	64.02SheetSage1	75.29
Rock CorpusMelody	Vocal pitch-class F1	49.19SheetSage1	67.08

SOTA counts refer to the leading scores against SheetSage1, Madmom, and the task-specific systems in this comparison. SheetSage2-AR uses one model selected by validation loss. Beat This! scores average three released training seeds. Melody F1 measures pitch-class notes with a 50-ms onset tolerance and no offset matching; structure F1 measures section boundaries at the stated tolerance.

On Chords1217, ChordFormer uses one held-out fold model per track; its per-track recall is weighted by reference span to match the other models’ scoring protocol. SheetSage1 and SheetSage2-AR each evaluate one fixed model on all 1,217 tracks. Madmom’s chord training data include Chords1217. JAAH includes all 113 recordings, with references intersected with each audio span. Rock Corpus reports the mean over all 200 recordings, including six empty references scored as zero.

Full comparison includes SheetSage1, Madmom, and task-specific systems.

Download all results

Training data

Our models are trained primarily on CC0 music and synthetic data. Tokenwave.AI provides most of our synthetic training data under license. We are committed to the ethical and responsible use of data.

MERT2: 700K hours
SheetSage2: 28.4K hours
YuE2: 346K hours

Citation

If you use YuE2, MERT2, SheetSage2, or WildSongBench, please cite our paper on arXiv (PDF).

BibTeX

@article{yuan2026yue2,
  title = {{YuE2}: Unifying Symbolic and Audio Music Generation at Frontier Quality},
  author = {Yuan, Ruibin and Pan, Jiahao and Jiang, Junyan and Wu, Zhiyue and Zhou, Ziya and Sun, Jiankai and Li, Yizhi and Zhang, Ge and Gu, Yicheng and Tian, Zeyue and Dai, Junyu and Lin, Hanfeng and Li, Kai and Wu, Shangda and Liu, Xuanjie and Wang, Jiaming and Liu, Zihan and Wang, Yue and Ma, Yinghao and Yin, Hanzhi and Chen, Kangrui and Zhang, Xinyue and Ma, Ziyang and Liao, Mengqi and Zhao, Hejia and Huang, Guowei and Yan, Chao and Ke, Lei and Yu, Jianwei and Liu, Bei and Guo, Joe and Xue, Liumeng and Xia, Gus and Xue, Wei and Guo, Yike},
  journal = {arXiv preprint arXiv:2609.33757},
  year = {2026},
  eprint = {2609.33757},
  archivePrefix = {arXiv},
  primaryClass = {eess.AS},
  url = {https://arxiv.org/abs/2609.33757}
}

Frontier quality. Music, with a plan.