Qwen 2.5 VL 3B Instruct: benchmark results
Provider: Alibaba. Released 2025-01-26. Access: Open.
Aggregate Rating 1486 ± 8, rank #2568 of 10916 rated models, from 511 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Bench2Drive-VL (BEV, CoT) | 48.05 | Driving Score (0-100; Bench2Drive closed-loop protocol: rout | 100 |
| Bench2Drive-VL (Camera, CoT) | 42.24 | Driving Score (0-100; Bench2Drive closed-loop protocol: rout | 100 |
| Bench2Drive-VL (Camera, No CoT) | 58.93 | Driving Score (0-100; Bench2Drive closed-loop protocol: rout | 100 |
| CHaystack - Gold-Image QA (Exact Match) | 35.44 | Exact match (%; 4,543 Chinese document questions (academic p | 100 |
| CHaystack - Gold-Image QA (F1) | 61.62 | Character-level F1 (%; 4,543 Chinese document questions (aca | 100 |
| DatBench - Document Understanding | 71.1 | Accuracy (%) | 100 |
| MiSCHiEF (Culture) - Image-to-Caption Matching | 87.46 | Accuracy (%, 0-100) on MiC: 279 culture minimal pairs, the s | 100 |
| DatBench - Spatial Reasoning | 27.5 | Accuracy (%) | 92.3 |
| NutriBench-Kitchen - Long-Term Planning (Hard) | 73.3 | Score (%; the long-term planning task family, hard regime: d | 90.9 |
| PanoSpace-Bench - Relative 3D Position | 35.2 | Accuracy (%; 250 four-option multiple-choice questions on on | 90.9 |
| PanoSpace-Bench - Seam Continuity | 46.4 | Accuracy (%; 250 two- or four-option multiple-choice questio | 90.9 |
| DatBench - Scene OCR | 75.1 | Accuracy (%) | 88.5 |
Interactive version: theaggregate.ai/model?slug=qwen-2-5-vl-3b-instruct · How It Works · Data refreshed daily, snapshot 2026-11-08.