GPT-4o: benchmark results
OpenAI GPT-4o omni model for general text, vision, and multimodal tasks. Provider: OpenAI. Released 2024-05-13. Access: API.
Aggregate Rating 1634 ± 2, rank #343 of 1656 rated models, from 4437 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| 360Bench (CMP) - Projection-Distorted Perception | 53.5 | Projection-distorted perception accuracy (%), mean of recogn | 100 |
| 360Bench (ERP) - Projection-Distorted Perception | 46.5 | Projection-distorted perception accuracy (%), mean of recogn | 100 |
| AGC-Bench - science_analogies | 2.5 | Dataset z-score | 100 |
| AIPatient Arena | 4.99 | Questioning Skills (QS) (self-reported) | 100 |
| ALARB | 46 | Facts+regulations correct (imported) | 100 |
| ARM-Thinker Study - ARMBench-VL Document Understanding | 58.7 | Judgment Accuracy (%) | 100 |
| ARM-Thinker Study - RewardBench-2 | 65.5 | Judgment Accuracy (%) | 100 |
| ARM-Thinker Study - VL-RewardBench Reasoning | 70.5 | Judgment Accuracy (%) | 100 |
| AbstentionBench - false premise - KUQ/False assumptions - F1 | 79.41 | Abstention F1 (%) | 100 |
| AbstentionBench - false premise - KUQ/False assumptions - Recall | 70.3 | Abstention Recall (%) | 100 |
| AbstentionBench - underspecified intent - KUQ/Ambiguous - F1 | 76.72 | Abstention F1 (%) | 100 |
| AbstentionBench - underspecified intent - KUQ/Ambiguous - Recall | 75.83 | Abstention Recall (%) | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-4o · How It Works · Data refreshed daily, snapshot 2026-10-20.