Phi-3-Vision: benchmark results
Provider: Microsoft. Released 2024-05-21. Access: Open.
Aggregate Rating 1414 ± 29, rank #5455 of 10916 rated models, from 445 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenVLM MMT-Bench - Traffic Participants Understanding | 70 | Score (%) | 97.1 |
| OpenVLM MMT-Bench - Transparent Object Detection | 85 | Score (%) | 96.6 |
| OpenVLM MMMU - Biology | 60 | Accuracy (%) | 93.5 |
| OpenVLM ChartQA Test - Augmented | 94.2 | Relaxed Accuracy (%) | 92.1 |
| OpenVLM MMT-Bench - Behavior Anomaly Detection | 65 | Score (%) | 91.5 |
| OpenVLM MMT-Bench - GUI Install | 55 | Score (%) | 90 |
| OpenVLM ChartQA Test | 81.8 | Relaxed Accuracy (%) | 87.9 |
| OpenVLM MMT-Bench - Video Captioning | 100 | Score (%) | 87.4 |
| OpenVLM MMT-Bench - Camouflage Object Detection | 65 | Score (%) | 86.4 |
| OpenVLM ChartQA Test - Human | 69.4 | Relaxed Accuracy (%) | 86 |
| OpenVLM MMT-Bench - Helmet Anomaly Detection | 90 | Score (%) | 85.9 |
| OpenVLM MMT-Bench - Chart To Text | 95 | Score (%) | 84 |
Interactive version: theaggregate.ai/model?slug=phi-3-vision · How It Works · Data refreshed daily, snapshot 2026-11-08.