GPT-4o Audio: benchmark results
Provider: OpenAI. Released 2024-10-17. Access: API.
Aggregate Rating 1590 ± 18, rank #1121 of 10916 rated models, from 168 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| APUN-Bench (Real-World) - Homographic Pun Inference | 50.7 | Homographic pun-inference accuracy (%), both predicted meani | 100 |
| APUN-Bench (Synthetic) - Coarse Pun Location | 75.97 | Coarse pun-location accuracy (%), the predicted pun word mat | 100 |
| APUN-Bench (Synthetic) - Fine Pun Location | 56.63 | Fine pun-location accuracy (%), the predicted pun word exact | 100 |
| APUN-Bench (Synthetic) - Heterographic Pun Inference | 59.06 | Heterographic pun-inference accuracy (%), the predicted alte | 100 |
| APUN-Bench (Synthetic) - Homographic Pun Inference | 57.81 | Homographic pun-inference accuracy (%), both predicted meani | 100 |
| Ambiguous Speech Emotion - CREMA-D | 0.29 | Jensen-Shannon divergence (0-1; between the emotion distribu | 100 |
| AudioBench - SLUE P2 SQA5 | 89.41 | Mean Score (%) | 100 |
| CASU - Contextual Reasoning | 74.02 | Accuracy (%; four-option multiple-choice questions on semi-s | 100 |
| CASU - Counterfactual Reasoning | 74.96 | Accuracy (%; four-option multiple-choice questions on semi-s | 100 |
| HumDial-EIBench - Text Empathy (Chinese, Gemini 2.5 Pro Judge) | 3.79 | Text empathy and insight score (1-5) of the model's empathet | 100 |
| HumDial-EIBench - Text Empathy (Chinese, Qwen3-Omni Judge) | 3.77 | Text empathy and insight score (1-5) of the model's empathet | 100 |
| HumDial-EIBench - Text Empathy (English, Gemini 2.5 Pro Judge) | 4.1 | Text empathy and insight score (1-5) of the model's empathet | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-4o-audio · How It Works · Data refreshed daily, snapshot 2026-11-08.