Qwen 2 VL 7B: benchmark results
Alibaba's Apache-2.0 7B vision-language model (August 2024) handling image, multi-image, and video input at a cost-effective size. Provider: Alibaba. Released 2024-08-29. Access: Open.
Aggregate Rating 1511 ± 6, rank #750 of 1656 rated models, from 1313 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| GeoMMBench - Spatial | 62.2 | Micro-averaged accuracy (%) on the spatial relation question | 100 |
| MEGA-Bench Task - VLN English Next Step | 40 | Task Score (%) | 100 |
| MEGA-Bench Task - Weather Info Parsing | 97.2 | Task Score (%) | 100 |
| OpenRef - Negative Rejection | 90.9 | N3R (0-100), negative relative rejection reliability: mean o | 100 |
| OpenVLM VCR | 75.6 | Overall Jaccard (%) | 100 |
| OpenVLM VCR - Chinese Easy | 78.4 | Jaccard (%) | 100 |
| OpenVLM VCR - Chinese Hard | 43.8 | Jaccard (%) | 100 |
| OpenVLM VCR - English Easy | 93.8 | Jaccard (%) | 100 |
| OpenVLM MMT-Bench - Traffic Sign Understanding | 100 | Score (%) | 99.8 |
| OpenVLM MMT-Bench - Religious Recognition | 90 | Score (%) | 99.5 |
| OpenVLM MMT-Bench - Pixel Localization | 60 | Score (%) | 99.3 |
| MEGA-Bench Task - Autorater Subject | 71.4 | Task Score (%) | 98.8 |
Interactive version: theaggregate.ai/model?slug=qwen-2-vl-7b · How It Works · Data refreshed daily, snapshot 2026-10-20.