GPT-4o-mini-Audio: benchmark results
Provider: OpenAI. Access: API.
Aggregate Rating 1539 ± 41, rank #1846 of 10916 rated models, from 34 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Jamendo-MT-QA - Yes-No Questions | 77.3 | Accuracy (%) on comparative yes/no questions about two music | 100 |
| PARSA-Bench - Reading Comprehension (TinyStories) | 80 | Accuracy (0-1 times 100) on 500 spoken Persian TinyStories p | 100 |
| IHBench - Recovery Quality - Normal | 83 | Recovery pass rate (%; 81 normal cut-ins, every type-specifi | 98.1 |
| IHBench - Recovery Quality - Topic Switch | 88 | Recovery pass rate (%; 73 topic switches, every type-specifi | 96.2 |
| IHBench - Recovery Quality - Impatient | 64 | Recovery pass rate (%; 84 impatient interruptions, every typ | 90.4 |
| PARSA-Bench - Intent Detection | 47 | Macro-F1 (0-1 times 100) over the union of gold and predicte | 71.4 |
| PARSA-Bench - Named Entity Recognition | 17 | Exact-match entity F1 (0-1 times 100) on 497 spoken Persian | 71.4 |
| PARSA-Bench - Reading Comprehension (Wikipedia) | 91 | Accuracy (0-1 times 100) on 500 spoken Wikipedia passages wi | 71.4 |
| IHBench - Recovery Quality | 65.4 | Recovery pass rate (%; share of 428 interruption points meet | 71.2 |
| MCGA - Speech Translation | 43.5 | LLM-B (0-100; mean of beauty of form, meaning and sound of t | 66.7 |
| PARSA-Bench - Multiple-Choice QA | 40 | Accuracy (0-1 times 100) on 500 spoken ParsiNLU four-option | 64.3 |
| IHBench - Recovery Quality - Pushback | 70 | Recovery pass rate (%; 105 pushbacks, every type-specific cr | 61.5 |
Interactive version: theaggregate.ai/model?slug=gpt-4o-mini-audio · How It Works · Data refreshed daily, snapshot 2026-11-08.