AudioJevBench
Models listen to a clip and assign probabilities to answers about speech or sound. The maker checks accuracy and calibration across 988 scored items.
Benchmark Heaven tested 17 systems. Nine completed both splits, six ran only on public items, and two sound classifiers covered one task family.
- Made by
- Benchmark Heaven
- Released
- September 2026
- Kind
- Audio understanding and voice
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- Public, plus a private test set
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- A host
- In views
- Audio and speech
- Last model added
- September 2025
- Scores refreshed
- November 20th, 2026
Qwen3-Omni-30B-A3B leads with 70.4 among models released by September 2025 Qwen3-Omni-30B-A3B leads with 70.4 among models released by September 2025
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. 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 September 2025, 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
Qwen3-Omni-30B-A3B self-hosted GPU 70.4
- 2
Qwen2.5-Omni-7B self-hosted GPU 52.1
- 3 Qwen2.5-Omni-3B self-hosted GPU 48.2
- 4 Voxtral Mini 3B (2507) self-hosted GPU 36.6
- 5 Whisper large-v3-turbo + qwen3-4b-instruct-2507 self-hosted GPU 19.3
- 6 MOSS-Audio-4B-Instruct self-hosted GPU 12.7
- 7 AudioJev (mocomoco edge ONNX) local CPU 0.00
- 8
Qwen2-Audio-7B-Instruct self-hosted GPU 0.00
- 9
Ultravox v0.5 Llama 3.2 1B self-hosted GPU 0.00
Source: Official AudioJevBench leaderboard
- 1
Qwen3-Omni-30B-A3B self-hosted GPU 70.4
- 2
Qwen2.5-Omni-7B self-hosted GPU 52.1
- 3
Qwen2-Audio-7B-Instruct self-hosted GPU 0.00
- 4
Ultravox v0.5 Llama 3.2 1B self-hosted GPU 0.00
Source: Official AudioJevBench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 70.4 | self-hosted GPU | Sep 2025 | Official leaderboard |
| 2 | | 52.1 | self-hosted GPU | Mar 2025 | Official leaderboard |
| 3 | Qwen2.5-Omni-3B | 48.2 | self-hosted GPU | Official leaderboard | |
| 4 | Voxtral Mini 3B (2507) | 36.6 | self-hosted GPU | Official leaderboard | |
| 5 | Whisper large-v3-turbo + qwen3-4b-instruct-2507 | 19.3 | self-hosted GPU | Official leaderboard | |
| 6 | MOSS-Audio-4B-Instruct | 12.7 | self-hosted GPU | Official leaderboard | |
| 7 | AudioJev (mocomoco edge ONNX) | 0.00 | local CPU | Official leaderboard | |
| 8 | | 0.00 | self-hosted GPU | Aug 2024 | Official leaderboard |
| 9 | | 0.00 | self-hosted GPU | Feb 2025 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 70.4 | self-hosted GPU | Sep 2025 | Official leaderboard |
| 2 | | 52.1 | self-hosted GPU | Mar 2025 | Official leaderboard |
| 3 | | 0.00 | self-hosted GPU | Aug 2024 | Official leaderboard |
| 4 | | 0.00 | self-hosted GPU | Feb 2025 | Official leaderboard |
Rankings by part
We read 12 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Clean audio, Noisy (≤ 5 dB SNR) and 15 dB. Qwen3-Omni-30B-A3B leads 10 of them, and Qwen2.5-Omni-7B leads the other 2. Arrows show how far a model moved from its overall rank among the models we track.
Clean audio
4 models we track have a score here.
- 1
Qwen3-Omni-30B-A3B 87.3%
- 2
Qwen2.5-Omni-7B 77.3%
- 3
Ultravox v0.5 Llama 3.2 1B ↑1 16%
- 4
Qwen2-Audio-7B-Instruct ↓1 10%
Noisy (≤ 5 dB SNR)
4 models we track have a score here.
- 1
Qwen3-Omni-30B-A3B 88.6%
- 2
Qwen2.5-Omni-7B 76.2%
- 3
Ultravox v0.5 Llama 3.2 1B ↑1 14.7%
- 4
Qwen2-Audio-7B-Instruct ↓1 6.08%
15 dB
4 models we track have a score here.
- 1
Qwen3-Omni-30B-A3B 92.5%
- 2
Qwen2.5-Omni-7B 82.9%
- 3
Ultravox v0.5 Llama 3.2 1B ↑1 18.4%
- 4
Qwen2-Audio-7B-Instruct ↓1 13.1%
10 dB
4 models we track have a score here.
- 1
Qwen3-Omni-30B-A3B 92.9%
- 2
Qwen2.5-Omni-7B 85.3%
- 3
Ultravox v0.5 Llama 3.2 1B ↑1 27.5%
- 4
Qwen2-Audio-7B-Instruct ↓1 14.4%
5 dB
4 models we track have a score here.
- 1
Qwen3-Omni-30B-A3B 89.6%
- 2
Qwen2.5-Omni-7B 74.2%
- 3
Ultravox v0.5 Llama 3.2 1B ↑1 16.4%
- 4
Qwen2-Audio-7B-Instruct ↓1 4.75%
0 dB
4 models we track have a score here.
- 1
Qwen3-Omni-30B-A3B 82.6%
- 2
Qwen2.5-Omni-7B 79.3%
- 3
Ultravox v0.5 Llama 3.2 1B ↑1 12.1%
- 4
Qwen2-Audio-7B-Instruct ↓1 5.43%
English
4 models we track have a score here.
- 1
Qwen3-Omni-30B-A3B 88.7%
- 2
Qwen2.5-Omni-7B 80.4%
- 3
Ultravox v0.5 Llama 3.2 1B ↑1 16.7%
- 4
Qwen2-Audio-7B-Instruct ↓1 12.8%
Non-English
4 models we track have a score here.
- 1
Qwen3-Omni-30B-A3B 87.9%
- 2
Qwen2.5-Omni-7B 74.7%
- 3
Ultravox v0.5 Llama 3.2 1B ↑1 17.8%
- 4
Qwen2-Audio-7B-Instruct ↓1 5.47%
Background talker
4 models we track have a score here.
- 1
Qwen3-Omni-30B-A3B 84.2%
- 2
Qwen2.5-Omni-7B 73.5%
- 3
Ultravox v0.5 Llama 3.2 1B ↑1 9.55%
- 4
Qwen2-Audio-7B-Instruct ↓1 6.75%
Overlapping talkers
4 models we track have a score here.
- 1
Qwen2.5-Omni-7B ↑1 75.9%
- 2
Qwen3-Omni-30B-A3B ↓1 74.1%
- 3
Qwen2-Audio-7B-Instruct 8.62%
- 4
Ultravox v0.5 Llama 3.2 1B 6.90%
Speaker verification
4 models we track have a score here.
- 1
Qwen3-Omni-30B-A3B 63.8%
- 2
Qwen2.5-Omni-7B 50.9%
- 3
Ultravox v0.5 Llama 3.2 1B ↑1 1.72%
- 4
Qwen2-Audio-7B-Instruct ↓1 0%
Sound events
4 models we track have a score here.
- 1
Qwen2.5-Omni-7B ↑1 90.2%
- 2
Qwen3-Omni-30B-A3B ↓1 89.8%
- 3
Ultravox v0.5 Llama 3.2 1B ↑1 19.1%
- 4
Qwen2-Audio-7B-Instruct ↓1 9.93%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official AudioJevBench 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 AudioJevBench 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
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 14 months ago, in September 2025, for Qwen3-Omni-30B-A3B. Every benchmark
Catalog record 2 pass
- Pass Task access Public tasks, plus a private test set.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (9 entries).
- Pass Models scored 4 models we track, plus 5 other entries on its leaderboard.
- Warning Newest score The newest came 14 months ago, in September 2025, for Qwen3-Omni-30B-A3B.
- Note Frontier models None of its models are on the capabilities index.
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on benchmarkheaven.com: 4 rows for models we track and 5 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.