Qwen2.5-VL Instruct (3B): benchmark results
Provider: Alibaba. Access: Open.
Aggregate Rating 1535 ± 23, rank #1897 of 10916 rated models, from 175 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HELM VHELM - Pope | 89.4 | PEM | 97.9 |
| HELM Robo-Reward-Bench - Berkeley Cable Routing | 0.96 | Absolute error | 94.7 |
| HELM Robo-Reward-Bench - Dlr Sara Grid Clamp Converted Externally To Rlds | 1.55 | Absolute error | 94.7 |
| HELM VHELM - Blink Perception - Visual Similarity | 97.78 | Prefix Quasi-Exact Match (%) | 92.6 |
| HELM VHELM - Blink Perception - Art Style | 97.44 | Prefix Quasi-Exact Match (%) | 89.4 |
| HELM VHELM - Mme - Landmark | 95.25 | Prefix Quasi-Exact Match (%) | 88.3 |
| HELM VHELM - Exams V - Chinese (Natural Science) | 62.11 | Prefix Quasi-Exact Match (%) | 83 |
| HELM VHELM - Blink Reasoning - IQ Test | 48.67 | Prefix Quasi-Exact Match (%) | 81.9 |
| HELM VHELM - Unicorn - OODCV-VQA | 81.7 | Prefix Quasi-Exact Match (%) | 78.7 |
| HELM VHELM - Vqa Base | 85 | PEM | 78.7 |
| HELM VHELM - Exams V - Bulgarian (Natural Science) | 14.81 | Prefix Quasi-Exact Match (%) | 75.5 |
| HELM VHELM - Viz Wiz | 76.7 | PEM | 74.5 |
Interactive version: theaggregate.ai/model?slug=qwen2-5-vl-instruct-3b · How It Works · Data refreshed daily, snapshot 2026-11-08.