HMU-Bench
Models answer Chinese and English questions about musical notation and recordings. The authors score responses against reference answers.
- Made by
- Zheqi Dai and colleagues
- Released
- October 2026
- Kind
- Audio understanding and voice
- Level
- University
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Subject
- Arts
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Vision, Audio and speech
- Strongest model tested
-
Gemini 3.7 Flash #26 on the capability index
- Last model added
- August 2026
- Scores refreshed
- November 20th, 2026
Gemini 3.7 Flash leads with 2.47 among models released by August 2026 Gemini 3.7 Flash leads with 2.47 among models released by August 2026
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. The headline is the plain mean of the Chinese and English scores for the ABC notation and audio tracks. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in August 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1
Gemini 3.7 Flash 2.47
- 2
Seed2.0 Lite (0215) 2.27
- 3
Gemini 3.1 Pro 2.20
- 4 StepAudio 3 Realtime 2.13
- 5
Qwen3.5-Omni-Plus 2.01
- 6 Gemini 2.5 Pro (snapshot unspecified) 1.93
- 7 Qwen 3 Omni Flash 1.42
Source: Official HMU-Bench leaderboard
- 1
Gemini 3.7 Flash 2.47
- 2
Seed2.0 Lite (0215) 2.27
- 3
Gemini 3.1 Pro 2.20
- 4
Qwen3.5-Omni-Plus 2.01
Source: Official HMU-Bench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 2.47 | Aug 2026 | Official leaderboard | |
| 2 | | 2.27 | Feb 2026 | Official leaderboard | |
| 3 | | 2.20 | Feb 2026 | Official leaderboard | |
| 4 | StepAudio 3 Realtime | 2.13 | Official leaderboard | ||
| 5 | | 2.01 | Mar 2026 | Official leaderboard | |
| 6 | Gemini 2.5 Pro (snapshot unspecified) | 1.93 | Official leaderboard | ||
| 7 | Qwen 3 Omni Flash | 1.42 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 2.47 | Aug 2026 | Official leaderboard | |
| 2 | | 2.27 | Feb 2026 | Official leaderboard | |
| 3 | | 2.20 | Feb 2026 | Official leaderboard | |
| 4 | | 2.01 | Mar 2026 | Official leaderboard |
Rankings by part
We read 23 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as ABC Chinese, ABC English and Audio Chinese. Gemini 3.7 Flash leads 13 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
ABC Chinese
4 models we track have a score here.
- 1
Gemini 3.7 Flash 3.84
- 2
Seed2.0 Lite (0215) 3.52
- 3
Gemini 3.1 Pro 3.27
- 4
Qwen3.5-Omni-Plus 2.74
ABC English
4 models we track have a score here.
- 1
Gemini 3.7 Flash 3.86
- 2
Seed2.0 Lite (0215) 3.52
- 3
Gemini 3.1 Pro 3.24
- 4
Qwen3.5-Omni-Plus 2.62
Audio Chinese
4 models we track have a score here.
- 1
Qwen3.5-Omni-Plus ↑3 1.33
- 2
Gemini 3.1 Pro ↑1 1.18
- 3
Gemini 3.7 Flash ↓2 1.16
- 4
Seed2.0 Lite (0215) ↓2 1.09
Audio English
4 models we track have a score here.
- 1
Qwen3.5-Omni-Plus ↑3 1.33
- 2
Gemini 3.1 Pro ↑1 1.12
- 3
Gemini 3.7 Flash ↓2 1.03
- 4
Seed2.0 Lite (0215) ↓2 0.94
global information
4 models we track have a score here.
- 1
Gemini 3.7 Flash 3.97
- 2
Seed2.0 Lite (0215) 3.79
- 3
Gemini 3.1 Pro 3.73
- 4
Qwen3.5-Omni-Plus 3.33
pitch and melody
4 models we track have a score here.
- 1
Gemini 3.7 Flash 3.69
- 2
Seed2.0 Lite (0215) 3.34
- 3
Gemini 3.1 Pro 2.95
- 4
Qwen3.5-Omni-Plus 2.40
tonality and harmony
4 models we track have a score here.
- 1
Gemini 3.7 Flash 3.91
- 2
Seed2.0 Lite (0215) 3.52
- 3
Gemini 3.1 Pro 3.45
- 4
Qwen3.5-Omni-Plus 2.72
rhythm
4 models we track have a score here.
- 1
Gemini 3.7 Flash 3.95
- 2
Seed2.0 Lite (0215) 3.72
- 3
Gemini 3.1 Pro 3.24
- 4
Qwen3.5-Omni-Plus 2.81
form and temporal structure
4 models we track have a score here.
- 1
Gemini 3.7 Flash 3.76
- 2
Seed2.0 Lite (0215) 3.40
- 3
Gemini 3.1 Pro 3.16
- 4
Qwen3.5-Omni-Plus 2.94
verifiable symbolic tasks
4 models we track have a score here.
- 1
Gemini 3.7 Flash 3.83
- 2
Seed2.0 Lite (0215) 3.47
- 3
Gemini 3.1 Pro 3.12
- 4
Qwen3.5-Omni-Plus 2.60
extraction
4 models we track have a score here.
- 1
Gemini 3.7 Flash 3.81
- 2
Seed2.0 Lite (0215) 3.56
- 3
Gemini 3.1 Pro 3.11
- 4
Qwen3.5-Omni-Plus 2.50
single step
4 models we track have a score here.
- 1
Gemini 3.7 Flash 3.86
- 2
Seed2.0 Lite (0215) 3.51
- 3
Gemini 3.1 Pro 3.26
- 4
Qwen3.5-Omni-Plus 2.77
multi step
4 models we track have a score here.
- 1
Gemini 3.7 Flash 3.89
- 2
Seed2.0 Lite (0215) 3.48
- 3
Gemini 3.1 Pro 3.44
- 4
Qwen3.5-Omni-Plus 2.77
global easy
4 models we track have a score here.
- 1
Qwen3.5-Omni-Plus ↑3 2.52
- 2
Gemini 3.1 Pro ↑1 2.25
- 3
Seed2.0 Lite (0215) ↓1 1.89
- 4
Gemini 3.7 Flash ↓3 1.86
global hard
4 models we track have a score here.
- 1
Qwen3.5-Omni-Plus ↑3 1.21
- 2
Gemini 3.1 Pro ↑1 0.92
- 3
Gemini 3.7 Flash ↓2 0.91
- 4
Seed2.0 Lite (0215) ↓2 0.81
local easy
4 models we track have a score here.
- 1
Qwen3.5-Omni-Plus ↑3 1.81
- 2
Gemini 3.1 Pro ↑1 1.55
- 3
Gemini 3.7 Flash ↓2 1.54
- 4
Seed2.0 Lite (0215) ↓2 1.51
local medium
4 models we track have a score here.
- 1
Gemini 3.7 Flash 1.00
- 2
Gemini 3.1 Pro ↑1 0.89
- 3
Qwen3.5-Omni-Plus ↑1 0.85
- 4
Seed2.0 Lite (0215) ↓2 0.77
local hard
4 models we track have a score here.
- 1
Qwen3.5-Omni-Plus ↑3 0.26
- 2
Gemini 3.7 Flash ↓1 0.15
- 3
Gemini 3.1 Pro 0.13
- 4
Seed2.0 Lite (0215) ↓2 0.10
global easy full-credit rate
4 models we track have a score here.
- 1
Qwen3.5-Omni-Plus ↑3 48.4%
- 2
Gemini 3.1 Pro ↑1 44%
- 3
Gemini 3.7 Flash ↓2 35.9%
- 4
Seed2.0 Lite (0215) ↓2 35.3%
global hard full-credit rate
4 models we track have a score here.
- 1
Qwen3.5-Omni-Plus ↑3 9.24%
- 2
Gemini 3.7 Flash ↓1 4.89%
- 3
Gemini 3.1 Pro 4.35%
- 4
Seed2.0 Lite (0215) ↓2 2.72%
local easy full-credit rate
4 models we track have a score here.
- 1
Qwen3.5-Omni-Plus ↑3 23.4%
- 2
Gemini 3.7 Flash ↓1 22.8%
- 3
Gemini 3.1 Pro 22.3%
- 4
Seed2.0 Lite (0215) ↓2 20.1%
local medium full-credit rate
4 models we track have a score here.
- 1
Gemini 3.7 Flash 10.9%
- 2
Gemini 3.1 Pro ↑1 10.3%
- 3
Seed2.0 Lite (0215) ↓1 7.61%
- 4
Qwen3.5-Omni-Plus 5.98%
local hard full-credit rate
4 models we track have a score here.
- 1
Gemini 3.1 Pro ↑2 0.54%
- 2
Gemini 3.7 Flash ↓1 0.54%
- 3
Seed2.0 Lite (0215) ↓1 0%
- 4
Qwen3.5-Omni-Plus 0%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official HMU-Bench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official HMU-Bench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Audio and video
- Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy.
Qwen 3.8 Max 88.7%
- LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization.
Gemini 4 Argon 91.7%
- MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy.
Qwen 3.8 Max 82.4%
- MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy.
Qwen 3.8 Max 90.8%
- FLEURS In use Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better.
Phi-4-multimodal 4.00%
- VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules.
Qwen3.5-Omni-Plus 93.1%
Data audit
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (7 entries).
- Pass Models scored 4 models we track, plus 3 other entries on its leaderboard.
- Pass Newest score The newest is from August 2026, for Gemini 3.7 Flash.
- Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 3.7 Flash (#26).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on github.com: 4 rows for models we track and 3 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.