Qwen 2.5 VL 7B: benchmark results
Alibaba's Qwen2.5-VL-7B, a vision-language model recognizing objects, charts, icons and layouts, acting as a visual agent for computer/phone use and understanding videos over an hour long. Provider: Alibaba. Released 2025-01-26. Access: Open.
Aggregate Rating 1534 ± 6, rank #1796 of 10870 rated models, from 1042 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| IndoTabVQA - Arabic | 23 | In-Match accuracy (%), the normalized gold answer found insi | 100 |
| SDGBiasBench - MCQ (No Context) | 53 | Accuracy (%; three-option ordinal questions on SDG indicator | 100 |
| SDGBiasBench - MCQ (Reduced Variable Names) | 53 | Accuracy (%; three-option ordinal questions on SDG indicator | 100 |
| Seeing Culture - Grounding mIoU | 47.2 | Overall mIoU (%) | 100 |
| OpenVLM SEEDBench IMG - Visual Reasoning | 84 | Accuracy (%) | 99.6 |
| OpenVLM MMT-Bench - Spot The Similarity | 100 | Score (%) | 99.3 |
| OpenVLM Q-Bench - Yes-or-No (In-context Distortion) | 82.9 | Accuracy (%) | 99.1 |
| OpenVLM MMT-Bench - Film and Television Recognition | 100 | Score (%) | 99 |
| OpenVLM MMT-Bench - Lesion Grading | 90 | Score (%) | 99 |
| OpenVLM MTVQA - Korean | 38 | Accuracy (%) | 98.7 |
| OpenVLM MTVQA - Vietnamese | 44.8 | Accuracy (%) | 98.7 |
| OpenVLM MMT-Bench - Industrial Produce Anomaly Detection | 90 | Score (%) | 98.3 |
Interactive version: theaggregate.ai/model?slug=qwen-2-5-vl-7b · How It Works · Data refreshed daily, snapshot 2026-11-12.