# Multimodal Audio Understanding on MMAU mini v05.15.25 (test)

Source: [Multimodal Audio Understanding on MMAU mini v05.15.25 (test)](<https://www.sota2.com/research/sota/multimodal-audio-understanding-on-mmau-mini-v05-15-25-test>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-11T09:04:00.289Z

Evaluation of Multimodal Audio Understanding capabilities using the test split of the MMAU mini dataset (version 05.15.25), reporting scores across Sound, Music, and Speech domains.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Multimodal Audio Understanding |
| Dataset (including any stated split) | MMAU mini v05.15.25 (test) |
| Metric | Sound Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 25 of 25 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Sound Accuracy](<https://www.sota2.com/research/sota/multimodal-audio-understanding-on-mmau-mini-v05-15-25-test.md?metric=Sound%20Accuracy>): Higher is better; unit: PERCENT.

- [Music Accuracy](<https://www.sota2.com/research/sota/multimodal-audio-understanding-on-mmau-mini-v05-15-25-test.md?metric=Music%20Accuracy>): Higher is better; unit: PERCENT.

- [Speech Accuracy](<https://www.sota2.com/research/sota/multimodal-audio-understanding-on-mmau-mini-v05-15-25-test.md?metric=Speech%20Accuracy>): Higher is better; unit: PERCENT.

- [Overall Accuracy](<https://www.sota2.com/research/sota/multimodal-audio-understanding-on-mmau-mini-v05-15-25-test.md?metric=Overall%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Sound Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| Audio Flamingo 3 | 79.6 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| Qwen2.5-Omni | 78.1 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| Audio Flamingo 2 | 71.5 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| DeSTA-2.5-Audio | 70.3 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| Audio Reasoner | 67.9 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| Qwen2-Audio-Instruct | 67.3 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| ALARM-CA | 66.4 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| Phi-4-Multimodal | 65.5 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| GPT-4o Audio | 64.6 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| ALARM-E | 64 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| AL-SSLAM-R | 61.3 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| ALARM-P | 58.6 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| Gemma 3n | 55.9 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| AL-Whisper-R | 52.3 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| Gemma 3n | 51.4 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| GPT-4o mini Audio | 50.8 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| AL-MuQ-R | 50.5 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| AL-W2VBERT2-R | 46.2 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| AL-Whisper-Instruct-R | 45.9 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| MusiLingo | 43.2 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| SALMONN | 41.1 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| MU-LLaMA | 33 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| GAMA | 31.8 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| GAMA-IT | 30.9 | Not recorded | 69b0f72f2d603e6de3e45fb9 |
| LTU | 20.4 | Not recorded | 69b0f72f2d603e6de3e45fb9 |

### Audio Flamingo 3

Sound Accuracy: 79.6. Result ID: 69b12ff96dbb7075b95bc71d.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a4d1e209131cea82504.md>)

Recorded evaluation configuration: {"Size":"8B","No LLM Modification":"×","# Samples":"26.7M","# Hours":"54.4K"}. This record does not assert independent reproduction.

### Qwen2.5-Omni

Sound Accuracy: 78.1. Result ID: 69b12ff96dbb7075b95bc71c.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77a21e209131cea7482c.md>)

Recorded evaluation configuration: {"Size":"7B","No LLM Modification":"×","# Tokens":"0.3T"}. This record does not assert independent reproduction.

### Audio Flamingo 2

Sound Accuracy: 71.5. Result ID: 69b12ff96dbb7075b95bc71f.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa774c1e209131cea7260a.md>)

Recorded evaluation configuration: {"Size":"3B","No LLM Modification":"×","# Samples":"5.9M"}. This record does not assert independent reproduction.

### DeSTA-2.5-Audio

Sound Accuracy: 70.3. Result ID: 69b12ff96dbb7075b95bc71a.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79d71e209131cea7fae2.md>)

Recorded evaluation configuration: {"Size":"8B","No LLM Modification":"✓","# Samples":"5M","# Hours":"7K"}. This record does not assert independent reproduction.

### Audio Reasoner

Sound Accuracy: 67.9. Result ID: 69b12ff96dbb7075b95bc71b.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b741e209131cea889a1.md>)

Recorded evaluation configuration: {"Size":"8B","No LLM Modification":"×","# Samples":"\> 1.2M","# Hours":"\> 320K"}. This record does not assert independent reproduction.

### Qwen2-Audio-Instruct

Sound Accuracy: 67.3. Result ID: 69b12ff96dbb7075b95bc719.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74281e209131cea6017e.md>)

Recorded evaluation configuration: {"Size":"7B","No LLM Modification":"×","# Hours":"320K"}. This record does not assert independent reproduction.

### ALARM-CA

Sound Accuracy: 66.4. Result ID: 69b12ff96dbb7075b95bc725.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

Recorded evaluation configuration: {"Size":"4B","No LLM Modification":"✓","# Samples":"5.5M","# Hours":"17K","# Tokens":"1.5B"}. This record does not assert independent reproduction.

### Phi-4-Multimodal

Sound Accuracy: 65.5. Result ID: 69b12ff96dbb7075b95bc71e.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77461e209131cea72272.md>)

Recorded evaluation configuration: {"Size":"4B","No LLM Modification":"×","# Samples":"\> 28M","# Hours":"\> 2M"}. This record does not assert independent reproduction.

### GPT-4o Audio

Sound Accuracy: 64.6. Result ID: 69b12ff96dbb7075b95bc710.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75681e209131cea67a98.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### ALARM-E

Sound Accuracy: 64. Result ID: 69b12ff96dbb7075b95bc727.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

Recorded evaluation configuration: {"Size":"4B","No LLM Modification":"✓","# Samples":"5.5M","# Hours":"17K","# Tokens":"1.5B"}. This record does not assert independent reproduction.

### AL-SSLAM-R

Sound Accuracy: 61.3. Result ID: 69b12ff96dbb7075b95bc723.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

Recorded evaluation configuration: {"Size":"4B","No LLM Modification":"✓","# Samples":"0.4M","# Hours":"1.2K","# Tokens":"108M"}. This record does not assert independent reproduction.

### ALARM-P

Sound Accuracy: 58.6. Result ID: 69b12ff96dbb7075b95bc726.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

Recorded evaluation configuration: {"Size":"4B","No LLM Modification":"✓","# Samples":"5.5M","# Hours":"17K","# Tokens":"1.5B"}. This record does not assert independent reproduction.

### Gemma 3n

Sound Accuracy: 55.9. Result ID: 69b12ff96dbb7075b95bc718.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

Recorded evaluation configuration: {"Size":"8B4E"}. This record does not assert independent reproduction.

### AL-Whisper-R

Sound Accuracy: 52.3. Result ID: 69b12ff96dbb7075b95bc724.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

Recorded evaluation configuration: {"Size":"4B","No LLM Modification":"✓","# Samples":"5.5M","# Hours":"17K","# Tokens":"1.5B"}. This record does not assert independent reproduction.

### Gemma 3n

Sound Accuracy: 51.4. Result ID: 69b12ff96dbb7075b95bc717.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

Recorded evaluation configuration: {"Size":"6B2E"}. This record does not assert independent reproduction.

### GPT-4o mini Audio

Sound Accuracy: 50.8. Result ID: 69b12ff96dbb7075b95bc70f.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75681e209131cea67a98.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### AL-MuQ-R

Sound Accuracy: 50.5. Result ID: 69b12ff96dbb7075b95bc722.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

Recorded evaluation configuration: {"Size":"4B","No LLM Modification":"✓","# Samples":"0.4M","# Hours":"1.5K","# Tokens":"137M"}. This record does not assert independent reproduction.

### AL-W2VBERT2-R

Sound Accuracy: 46.2. Result ID: 69b12ff96dbb7075b95bc721.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

Recorded evaluation configuration: {"Size":"4B","No LLM Modification":"✓","# Samples":"0.5M","# Hours":"0.6K","# Tokens":"58M"}. This record does not assert independent reproduction.

### AL-Whisper-Instruct-R

Sound Accuracy: 45.9. Result ID: 69b12ff96dbb7075b95bc720.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

Recorded evaluation configuration: {"Size":"4B","No LLM Modification":"✓","# Samples":"0.5M","# Hours":"1K","# Tokens":"91M"}. This record does not assert independent reproduction.

### MusiLingo

Sound Accuracy: 43.2. Result ID: 69b12ff96dbb7075b95bc716.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72c01e209131cea58630.md>)

Recorded evaluation configuration: {"Size":"7B","No LLM Modification":"✓","# Samples":"0.7M"}. This record does not assert independent reproduction.

### SALMONN

Sound Accuracy: 41.1. Result ID: 69b12ff96dbb7075b95bc715.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72d71e209131cea58d8a.md>)

Recorded evaluation configuration: {"Size":"13B","No LLM Modification":"×","# Samples":"2.3M","# Hours":"4.4K"}. This record does not assert independent reproduction.

### MU-LLaMA

Sound Accuracy: 33. Result ID: 69b12ff96dbb7075b95bc714.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70631e209131cea4ad62.md>)

Recorded evaluation configuration: {"Size":"7B","No LLM Modification":"✓","# Samples":"0.1M","# Hours":"76"}. This record does not assert independent reproduction.

### GAMA

Sound Accuracy: 31.8. Result ID: 69b12ff96dbb7075b95bc712.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73a71e209131cea5db04.md>)

Recorded evaluation configuration: {"Size":"7B","No LLM Modification":"×","# Samples":"6.3M"}. This record does not assert independent reproduction.

### GAMA-IT

Sound Accuracy: 30.9. Result ID: 69b12ff96dbb7075b95bc713.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73a71e209131cea5db04.md>)

Recorded evaluation configuration: {"Size":"7B","No LLM Modification":"×","# Samples":"6.5M"}. This record does not assert independent reproduction.

### LTU

Sound Accuracy: 20.4. Result ID: 69b12ff96dbb7075b95bc711.

[Reporting paper](<https://www.sota2.com/research/paper/69b0f72f2d603e6de3e45fb9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72431e209131cea555aa.md>)

[Implementation](<https://github.com/YuanGongND/ltu>)

Recorded evaluation configuration: {"Size":"7B","No LLM Modification":"×","# Samples":"5.6M"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
