MISHAP-Bench
Models answer open-ended questions about audio clips. A model judge uses human annotations to identify claims the recording does not support.
- Made by
- Zhi Wen Soi and colleagues
- Released
- September 2026
- Kind
- Audio understanding and voice
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Propensity
- Behavior
- Hallucination
- Language
- English
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- 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
Kimi-Audio-7B-Instruct leads with 32.8% among models released by August 2026 Kimi-Audio-7B-Instruct leads with 32.8% among models released by August 2026
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. 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 Kimi-Audio-7B-Instruct AAD 32.8%
- 2
Qwen2.5-Omni-7B MTI 35.7%
- 3
Gemini 3.7 Flash W/O 36.5%
- 4 Covo-Audio AAD 43.6%
- 5 MiMo-V2.5 W/O 47.5%
- 6
Qwen2-Audio-7B-Instruct VCD 48.3%
- 7
Audio Flamingo 3 AAD 48.9%
- 8
Qwen3-Omni-30B-A3B AAD 54.3%
- 9 MiMo-Audio-7B-Instruct AAD 56%
- 10 Step-Audio 2 mini DoLa 66.7%
Source: The MISHAP-Bench paper
- 1 Kimi-Audio-7B-Instruct AAD 32.8%
- 2
Qwen2.5-Omni-7B MTI 35.7%
- 3
Gemini 3.7 Flash W/O 36.5%
- 4 MiMo-V2.5 W/O 47.5%
- 5
Qwen2-Audio-7B-Instruct VCD 48.3%
- 6
Audio Flamingo 3 AAD 48.9%
- 7
Qwen3-Omni-30B-A3B AAD 54.3%
- 8 MiMo-Audio-7B-Instruct AAD 56%
Source: The MISHAP-Bench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Kimi-Audio-7B-Instruct | 32.8% | AAD | Apr 2025 | Official leaderboard |
| 2 | | 35.7% | MTI | Mar 2025 | Official leaderboard |
| 3 | | 36.5% | W/O | Aug 2026 | Official leaderboard |
| 4 | Covo-Audio | 43.6% | AAD | Official leaderboard | |
| 5 | MiMo-V2.5 | 47.5% | W/O | Apr 2026 | Official leaderboard |
| 6 | | 48.3% | VCD | Aug 2024 | Official leaderboard |
| 7 | | 48.9% | AAD | Jul 2025 | Official leaderboard |
| 8 | | 54.3% | AAD | Sep 2025 | Official leaderboard |
| 9 | MiMo-Audio-7B-Instruct | 56% | AAD | Sep 2025 | Official leaderboard |
| 10 | Step-Audio 2 mini | 66.7% | DoLa | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Kimi-Audio-7B-Instruct | 32.8% | AAD | Apr 2025 | Official leaderboard |
| 2 | | 35.7% | MTI | Mar 2025 | Official leaderboard |
| 3 | | 36.5% | W/O | Aug 2026 | Official leaderboard |
| 4 | MiMo-V2.5 | 47.5% | W/O | Apr 2026 | Official leaderboard |
| 5 | | 48.3% | VCD | Aug 2024 | Official leaderboard |
| 6 | | 48.9% | AAD | Jul 2025 | Official leaderboard |
| 7 | | 54.3% | AAD | Sep 2025 | Official leaderboard |
| 8 | MiMo-Audio-7B-Instruct | 56% | AAD | Sep 2025 | Official leaderboard |
Rankings by part
We read 18 parts from the benchmark’s paper in 5 groups: music, sound, speech, context and knowledge. Each ranks models on one part of the test. Kimi-Audio-7B-Instruct leads 7 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.
Context
8 models we track have a score here.
- 1
Qwen2.5-Omni-7B ↑1 50.3%
- 2 Kimi-Audio-7B-Instruct ↓1 50.6%
- 3
Audio Flamingo 3 ↑3 65.1%
- 4
Qwen2-Audio-7B-Instruct ↑1 67%
- 5 MiMo-Audio-7B-Instruct ↑3 67%
- 6 MiMo-V2.5 ↓2 72.1%
- 7
Gemini 3.7 Flash ↓4 73.6%
- 8
Qwen3-Omni-30B-A3B ↓1 74.6%
Knowledge
8 models we track have a score here.
- 1
Qwen2.5-Omni-7B ↑1 35.1%
- 2
Gemini 3.7 Flash ↑1 35.2%
- 3 Kimi-Audio-7B-Instruct ↓2 39.8%
- 4 MiMo-V2.5 45.5%
- 5
Qwen2-Audio-7B-Instruct 57.1%
- 6 MiMo-Audio-7B-Instruct ↑2 61.7%
- 7
Qwen3-Omni-30B-A3B 67.7%
- 8
Audio Flamingo 3 ↓2 68.9%
Overall
8 models we track have a score here.
- 1
Qwen2.5-Omni-7B ↑1 43.2%
- 2 Kimi-Audio-7B-Instruct ↓1 45.2%
- 3
Gemini 3.7 Flash 54.4%
- 4 MiMo-V2.5 58.8%
- 5
Qwen2-Audio-7B-Instruct 62.8%
- 6 MiMo-Audio-7B-Instruct ↑2 64.4%
- 7
Audio Flamingo 3 ↓1 67%
- 8
Qwen3-Omni-30B-A3B ↓1 74.9%
Context
8 models we track have a score here.
- 1 Kimi-Audio-7B-Instruct 49.5%
- 2
Qwen2.5-Omni-7B 50%
- 3
Audio Flamingo 3 ↑3 53.4%
- 4
Gemini 3.7 Flash ↓1 62.6%
- 5 MiMo-V2.5 ↓1 65.7%
- 6
Qwen3-Omni-30B-A3B ↑1 67.8%
- 7
Qwen2-Audio-7B-Instruct ↓2 68.1%
- 8 MiMo-Audio-7B-Instruct 72.4%
Knowledge
8 models we track have a score here.
- 1
Qwen2.5-Omni-7B ↑1 15.2%
- 2 Kimi-Audio-7B-Instruct ↓1 15.5%
- 3
Gemini 3.7 Flash 18.3%
- 4 MiMo-V2.5 28.2%
- 5
Audio Flamingo 3 ↑1 34.2%
- 6 MiMo-Audio-7B-Instruct ↑2 36.7%
- 7
Qwen2-Audio-7B-Instruct ↓2 44.7%
- 8
Qwen3-Omni-30B-A3B ↓1 44.8%
Overall
8 models we track have a score here.
- 1 Kimi-Audio-7B-Instruct 32.5%
- 2
Qwen2.5-Omni-7B 33%
- 3
Gemini 3.7 Flash 40.4%
- 4
Audio Flamingo 3 ↑2 43.8%
- 5 MiMo-V2.5 ↓1 46.9%
- 6 MiMo-Audio-7B-Instruct ↑2 54.5%
- 7
Qwen2-Audio-7B-Instruct ↓2 56.4%
- 8
Qwen3-Omni-30B-A3B ↓1 57.9%
Context
8 models we track have a score here.
- 1
Gemini 3.7 Flash ↑2 9.50%
- 2
Qwen2-Audio-7B-Instruct ↑3 20.6%
- 3
Qwen3-Omni-30B-A3B ↑4 22.5%
- 4 Kimi-Audio-7B-Instruct ↓3 23.4%
- 5
Qwen2.5-Omni-7B ↓3 25.2%
- 6
Audio Flamingo 3 27.1%
- 7 MiMo-V2.5 ↓3 28.6%
- 8 MiMo-Audio-7B-Instruct 51%
Knowledge
8 models we track have a score here.
- 1 Kimi-Audio-7B-Instruct 17.9%
- 2
Qwen2.5-Omni-7B 18.1%
- 3
Gemini 3.7 Flash 19.9%
- 4
Qwen2-Audio-7B-Instruct ↑1 23.4%
- 5
Qwen3-Omni-30B-A3B ↑2 37.1%
- 6
Audio Flamingo 3 44.9%
- 7 MiMo-V2.5 ↓3 45%
- 8 MiMo-Audio-7B-Instruct 47%
Overall
8 models we track have a score here.
- 1
Gemini 3.7 Flash ↑2 14.7%
- 2 Kimi-Audio-7B-Instruct ↓1 20.6%
- 3
Qwen2.5-Omni-7B ↓1 22.7%
- 4
Qwen2-Audio-7B-Instruct ↑1 23.6%
- 5
Qwen3-Omni-30B-A3B ↑2 29.8%
- 6
Audio Flamingo 3 36%
- 7 MiMo-V2.5 ↓3 36.8%
- 8 MiMo-Audio-7B-Instruct 49%
Contradicted Premise
8 models we track have a score here.
- 1 Kimi-Audio-7B-Instruct 44.5%
- 2
Qwen2.5-Omni-7B 49.3%
- 3
Audio Flamingo 3 ↑3 52.5%
- 4
Gemini 3.7 Flash ↓1 55.9%
- 5 MiMo-V2.5 ↓1 60.5%
- 6
Qwen3-Omni-30B-A3B ↑1 60.9%
- 7
Qwen2-Audio-7B-Instruct ↓2 61%
- 8 MiMo-Audio-7B-Instruct 69.8%
Absent Premise
8 models we track have a score here.
- 1
Gemini 3.7 Flash ↑2 34.8%
- 2
Qwen2.5-Omni-7B 38.4%
- 3 Kimi-Audio-7B-Instruct ↓2 42%
- 4
Qwen2-Audio-7B-Instruct ↑1 48.1%
- 5
Qwen3-Omni-30B-A3B ↑2 48.4%
- 6 MiMo-V2.5 ↓2 48.5%
- 7
Audio Flamingo 3 ↓1 48.7%
- 8 MiMo-Audio-7B-Instruct 71.9%
False Negation
8 models we track have a score here.
- 1 Kimi-Audio-7B-Instruct 31.5%
- 2
Qwen2-Audio-7B-Instruct ↑3 35.8%
- 3
Qwen2.5-Omni-7B ↓1 36.1%
- 4
Audio Flamingo 3 ↑2 37.5%
- 5 MiMo-Audio-7B-Instruct ↑3 39.2%
- 6
Gemini 3.7 Flash ↓3 40.7%
- 7
Qwen3-Omni-30B-A3B 44.5%
- 8 MiMo-V2.5 ↓4 47.8%
Overall
8 models we track have a score here.
- 1 Kimi-Audio-7B-Instruct 41.2%
- 2
Qwen2.5-Omni-7B 47.3%
- 3
Audio Flamingo 3 ↑3 48.5%
- 4
Gemini 3.7 Flash ↓1 48.6%
- 5
Qwen2-Audio-7B-Instruct 53.5%
- 6
Qwen3-Omni-30B-A3B ↑1 55%
- 7 MiMo-V2.5 ↓3 55.5%
- 8 MiMo-Audio-7B-Instruct 63.5%
Niche Specifics
8 models we track have a score here.
- 1 Kimi-Audio-7B-Instruct 22.4%
- 2
Gemini 3.7 Flash ↑1 22.5%
- 3
Qwen2.5-Omni-7B ↓1 25.8%
- 4
Qwen3-Omni-30B-A3B ↑3 34.7%
- 5 MiMo-V2.5 ↓1 35.8%
- 6 MiMo-Audio-7B-Instruct ↑2 41.4%
- 7
Qwen2-Audio-7B-Instruct ↓2 42.6%
- 8
Audio Flamingo 3 ↓2 43.4%
Unknowable Provenance
8 models we track have a score here.
- 1
Qwen2.5-Omni-7B ↑1 1.20%
- 2 Kimi-Audio-7B-Instruct ↓1 7.10%
- 3
Audio Flamingo 3 ↑3 17%
- 4 MiMo-Audio-7B-Instruct ↑4 23.3%
- 5
Qwen2-Audio-7B-Instruct 27%
- 6
Gemini 3.7 Flash ↓3 34.2%
- 7
Qwen3-Omni-30B-A3B 44.4%
- 8 MiMo-V2.5 ↓4 48.9%
False Authority
8 models we track have a score here.
- 1
Gemini 3.7 Flash ↑2 12.5%
- 2
Qwen2.5-Omni-7B 19.3%
- 3 MiMo-V2.5 ↑1 28.2%
- 4 Kimi-Audio-7B-Instruct ↓3 28.3%
- 5
Qwen2-Audio-7B-Instruct 49.1%
- 6
Qwen3-Omni-30B-A3B ↑1 59.9%
- 7 MiMo-Audio-7B-Instruct ↑1 72.6%
- 8
Audio Flamingo 3 ↓2 76%
False Premise
8 models we track have a score here.
- 1
Gemini 3.7 Flash ↑2 25.8%
- 2 Kimi-Audio-7B-Instruct ↓1 37.6%
- 3 MiMo-V2.5 ↑1 42.1%
- 4
Qwen2-Audio-7B-Instruct ↑1 46%
- 5
Qwen2.5-Omni-7B ↓3 48.7%
- 6 MiMo-Audio-7B-Instruct ↑2 59.2%
- 7
Qwen3-Omni-30B-A3B 59.7%
- 8
Audio Flamingo 3 ↓2 65.7%
Overall
8 models we track have a score here.
- 1
Qwen2.5-Omni-7B ↑1 23.8%
- 2 Kimi-Audio-7B-Instruct ↓1 24.4%
- 3
Gemini 3.7 Flash 24.4%
- 4 MiMo-V2.5 39.5%
- 5
Qwen2-Audio-7B-Instruct 42.2%
- 6 MiMo-Audio-7B-Instruct ↑2 48.5%
- 7
Audio Flamingo 3 ↓1 49.4%
- 8
Qwen3-Omni-30B-A3B ↓1 53.7%
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: The MISHAP-Bench paper
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: The MISHAP-Bench paper
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The MISHAP-Bench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official MISHAP-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 benchmark’s paper (10 entries).
- Pass Models scored 8 models we track, plus 2 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 Lower is better, so there’s no ceiling to reach.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 32 rows for models we track and 2 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.