Audio Understanding Benchmark (Mistral)
Voxtral Small 86.6% Dropped Models answer hard questions about 40 long-form audio recordings. Mistral built it in-house to test audio understanding beyond transcription.
Mistral reported it once, as a chart without printed values, in the Voxtral launch post. The grading method is not described, so model-judged is assumed.
- Level
- Expert Specialists with years of training, often over hours of work
- Scored
- Graded by a model A language model scores each response, usually against a rubric.
- Answers
- Short answer A number, word or short string matched to a key.
Voxtral Small leads with 86.6%
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
- 1 Voxtral Small 86.6%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Voxtral Small | 86.6% | Jul 2025 | Model card |
The best score passed 85% in Jul 2025
All the scores we have, one dot each, so a model tested at several settings appears more than once.
No launch post has listed it since Jul 2025
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| 15 Jul 2025 | Voxtral Small | AU Benchmark, 40 long-form audio examples (chart) |
One system card reports it
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 17 Jul 2025 | Voxtral Small technical report | 86.6 | AU Bench, internal speech understanding benchmark, LLM judge score (Table 8). SFT default checkpoint 86.61 with grade 4.16 of 5 (Table 2) | Capabilities |
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%
- 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%
- TVBench In use Models answer multiple-choice questions on 10 temporal video tasks, such as action counting and ordering, that static frames cannot solve. Scored by accuracy. Qwen 3.8 Max 81.9%