Qwen 3 VL 2B Instruct: benchmark results
Can locate interface elements for GUI agents, read text in images and generate code from visual layouts. Video training aligns events with timestamps for locating moments in a clip. Provider: Alibaba. Released 2025-10-21. Access: Open.
Aggregate Rating 1504 ± 10, rank #2283 of 10916 rated models, from 346 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| PRISM-VLM - Sycophancy Resistance | 89 | Per-axis mean score (%; share of turn-0-correct answers held | 100 |
| M-JudgeBench - Long CoT Same-Model Pairs | 60.53 | pairwise accuracy (%) on the 337 pairs where a correct and a | 92.1 |
| DatBench - Scene OCR | 72.3 | Accuracy (%) | 80.8 |
| DatBench - Document Understanding | 62.7 | Accuracy (%) | 80 |
| MemLens (32K) - Temporal Reasoning | 50.52 | Accuracy (%) on the temporal reasoning (TR, 194 questions) a | 73.1 |
| MemLens (64K) - Multi-Session Reasoning | 21.68 | Accuracy (%) on the multi-session reasoning (MSR, 143 questi | 73.1 |
| OddGridBench - Position | 7 | Accuracy (%) on the 200 images in which the odd item differs | 72.2 |
| PRISM-VLM | 56.9 | PScore (%; harmonic mean of the seven per-axis mean scores, | 67.5 |
| MemLens (128K) - Multi-Session Reasoning | 18.88 | Accuracy (%) on the multi-session reasoning (MSR, 143 questi | 66.7 |
| VSTAT - Atomic | 32.5 | Score (%) on questions about atomic state structures: 1,500 | 65.9 |
| DatBench - Math & Logic | 19.5 | Accuracy (%) | 65.4 |
| MERA Multi - RealVQA | 20.4 | Score (%) | 64.4 |
Interactive version: theaggregate.ai/model?slug=qwen-3-vl-2b-instruct · How It Works · Data refreshed daily, snapshot 2026-11-08.