Qwen 2 VL 2B: benchmark results
Alibaba's 2B instruction-tuned vision-language model in the Qwen2-VL series, handling images at varied resolutions and video understanding beyond 20 minutes, with multilingual text recognition. Provider: Alibaba. Released 2024-08-29. Access: Open.
Aggregate Rating 1402 ± 9, rank #1220 of 1656 rated models, from 1141 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenVLM MMT-Bench - Artwork Emotion Recognition | 70 | Score (%) | 100 |
| OpenVLM MMBench V1.1 EN - Social Relation | 93.4 | Accuracy (%) | 99.3 |
| OpenVLM MMT-Bench - 3D CAD Recognition | 80 | Score (%) | 98.5 |
| MEGA-Bench Task - MFC Bench Check Face Swap | 71.4 | Task Score (%) | 98.1 |
| OpenVLM MMT-Bench - AR | 59 | Score (%) | 98.1 |
| OpenVLM MMT-Bench - Counting By Visual Prompting | 55 | Score (%) | 98.1 |
| OpenVLM VCR - Chinese Easy | 61.5 | Jaccard (%) | 98 |
| OpenVLM VCR - Chinese Hard | 26.6 | Jaccard (%) | 98 |
| OpenVLM VCR | 61.3 | Overall Jaccard (%) | 97.9 |
| OpenVLM A-Bench - Nouns as Adjectives | 86 | Accuracy (%) | 97.2 |
| OpenVLM MMT-Bench - Geometrical Perspective | 91.7 | Score (%) | 96.6 |
| OpenVLM MMBench V1.1 EN - Image Topic | 98.9 | Accuracy (%) | 95.2 |
Interactive version: theaggregate.ai/model?slug=qwen-2-vl-2b · How It Works · Data refreshed daily, snapshot 2026-10-20.