Qwen 3 VL 2B: benchmark results
Provider: Alibaba. Released 2025-10-21.
Aggregate Rating 1511 ± 14, rank #2184 of 10916 rated models, from 242 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| COMEX - Crop mIoU | 0.54 | Mean crop IoU (0-1, native; overlap between the predicted cr | 100 |
| VLRS-Bench - Spatiotemporal Causal-Chain Reasoning | 42.8 | Score (%) on the 125 Spatiotemporal Causal-Chain Reasoning ( | 92.1 |
| TRAP - Image Privacy | 29.5 | Privacy score (%, 1 minus leakage) on the 200 image-document | 90.5 |
| PlantXpert - Visual Challenges | 77.16 | Accuracy (%) on the visual recognition questions on the expe | 90 |
| VLRS-Bench - Spatiotemporal Evolution Reasoning | 41.2 | Score (%) on the 125 Spatiotemporal Evolution Reasoning (two | 89.5 |
| CCTVBench - Balanced Accuracy | 62.96 | Balanced accuracy (%) over all binary video-question instanc | 88.9 |
| CoLT-Drive - False Positive | 91 | Pair accuracy (%; the 435 v_full samples, naturalistic surro | 87.5 |
| JaWildText - Handwriting OCR | 76 | Handwriting OCR score (0-1, scaled to 0-100): max(0, 1 - cha | 87.5 |
| DocAttriBench - Answer Locating | 25.3 | Box F1 (%; answer locating: given only the question, predict | 86.7 |
| EgoDyn-Bench - Temporal F1 | 37.5 | Macro-F1 (%) of temporal ego-motion questions (how the vehic | 86.7 |
| OpenRef - Celebrity | 89.1 | Grounding F1 (%) on proper-noun celebrity expressions, zero- | 83.3 |
| Sens-VisualNews - Strict | 92.7 | Top-1 accuracy (%) on the test split of the strict subset (s | 81.8 |
Interactive version: theaggregate.ai/model?slug=qwen-3-vl-2b · How It Works · Data refreshed daily, snapshot 2026-11-08.