Qwen2.5-VL Instruct (72B): benchmark results
Provider: Alibaba. Access: Open.
Aggregate Rating 1676 ± 22, rank #469 of 10916 rated models, from 174 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HELM VHELM - Exams V - Chinese (Natural Science) | 80.35 | Prefix Quasi-Exact Match (%) | 100 |
| HELM VHELM - Exams V - Chinese (Social Sciences) | 87.89 | Prefix Quasi-Exact Match (%) | 100 |
| HELM VHELM - Mmmu - Geography | 81.48 | Prefix Quasi-Exact Match (%) | 98.9 |
| HELM VHELM - Crossmodal 3600 - Hindi Locale | 3.74 | Prometheus Vision Rating (1-5) | 97.9 |
| HELM VHELM - Hateful Memes | 75.82 | PEM | 97.9 |
| HELM VHELM - Knowledge | 86.38 | Mean win rate | 97.9 |
| HELM VHELM - Toxicity | 97.87 | Mean win rate | 97.9 |
| HELM Robo-Reward-Bench - Berkeley Mvp Converted Externally To Rlds | 0.44 | Absolute Error | 95.8 |
| HELM VHELM - A Okvqa Dialect | 90.3 | PEM (Fairness) | 95.7 |
| HELM VHELM - Core Scenarios | 77.99 | Mean win rate | 95.7 |
| HELM VHELM - Mm Star Perception | 75.4 | PEM | 95.7 |
| HELM VHELM - Mm Star Perception - Fine-grained Perception | 69.6 | Prefix Quasi-Exact Match (%) | 95.7 |
Interactive version: theaggregate.ai/model?slug=qwen2-5-vl-instruct-72b · How It Works · Data refreshed daily, snapshot 2026-11-08.