Phi-3-Vision-128K-Instruct: benchmark results
Provider: Microsoft. Access: Open.
Aggregate Rating 1348 ± 38, rank #7762 of 10916 rated models, from 49 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ThermEval - Thermal Reasoning (Two People) | 56 | Accuracy (times 100, 0-100) on Task 5, whether the chest, fo | 87 |
| ThermEval - Temperature Estimation at 2 ft | 1.19 | Mean absolute error (degrees Celsius, 0 or more) on Task 7, | 84.1 |
| DisasterBench (UAV) - Propagation Path Inference | 74.7 | Accuracy (%) on propagation path inference (PPI) questions o | 80 |
| ThermEval - Colorbar Detection | 100 | Accuracy (times 100, 0-100) on Task 4, whether a ThermEval-D | 75 |
| DisasterBench (UAV) - Disaster-Chain Summary | 83.84 | Accuracy (%) on disaster-chain summary (DCS) questions on th | 65 |
| GeoMMBench - GIS | 63.6 | Micro-averaged accuracy (%) on the geographic information sy | 64.7 |
| ThermEval - Temperature Estimation (Coordinates) | 3.49 | Mean absolute error (degrees Celsius, 0 or more) on Task 6, | 60.9 |
| ThermEval - Human Counting (FLIR) | 3.59 | Mean absolute error (people, 0 or more) on Task 3, the numbe | 58.3 |
| GeoMMBench - SAR | 44 | Micro-averaged accuracy (%) on the synthetic aperture radar | 55.9 |
| DisasterBench (UAV) - Cascading Risk Reasoning | 67.19 | Accuracy (%) on cascading risk reasoning (CRR) questions on | 55 |
| ThermEval - Temperature Estimation (Region) | 2.26 | Mean absolute error (degrees Celsius, 0 or more) on Task 6, | 54.5 |
| ThermEval - Temperature Estimation at 6 ft | 1.14 | Mean absolute error (degrees Celsius, 0 or more) on Task 7, | 54.5 |
Interactive version: theaggregate.ai/model?slug=phi-3-vision-128k-instruct · How It Works · Data refreshed daily, snapshot 2026-11-08.