Qwen 2.5 VL 32B: benchmark results
Provider: Alibaba. Released 2025-03-24. Access: Open.
Aggregate Rating 1605 ± 11, rank #969 of 10916 rated models, from 307 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Creation-MMBench | 6.97 | Overall Reward | 100 |
| MMAlignBench | 75 | Overall Reward | 100 |
| MedRepBench - Explanation Acceptability | 67.83 | Acceptability rate (%; share of patient-facing explanations | 100 |
| MedRepBench - Item Name Recall | 90.13 | Field-level recall (%; item name of each ground-truth lab it | 100 |
| MedRepBench - Reference Range Recall | 81.61 | Field-level recall (%; reference range of each ground-truth | 100 |
| MedRepBench - Unit Recall | 79.3 | Field-level recall (%; unit of each ground-truth lab item, m | 100 |
| MedRepBench - Value Recall | 76.32 | Field-level recall (%; measured value of each ground-truth l | 100 |
| OpenRef - Celebrity | 91.4 | Grounding F1 (%) on proper-noun celebrity expressions, zero- | 100 |
| OpenRef - Multi Target | 37.4 | Grounding F1 (%) on expressions that refer to several object | 100 |
| OpenRef - Ordinal | 46.7 | Grounding F1 (%) on expressions with ordinal terms, zero-sho | 100 |
| OpenVLM MMBench V1.1 EN - Image Scene | 92.4 | Accuracy (%) | 98.4 |
| OpenVLM MMBench V1.1 EN - Structuralized Imagetext Understanding | 88.1 | Accuracy (%) | 98.4 |
Interactive version: theaggregate.ai/model?slug=qwen-2-5-vl-32b · How It Works · Data refreshed daily, snapshot 2026-11-08.