Qwen 3 VL 2B Instruct — benchmark results
Alibaba's Apache-2.0 2B dense Qwen3-VL instruct model (October 2025), the family's smallest vision-language variant, aimed at edge devices. Provider: Alibaba. Released 2025-10-19. Access: Open.
Unified ELO 1385 ± 35, rank #1301 of 1839 rated models, from 17 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| UGI - Willingness (W/10) | 6.5 | W/10 Score | 62.1 |
| TriViewBench | 44.07 | Overall (self-reported) | 55.6 |
| WikiVQABench | 56.4 | Accuracy (self-reported) | 50 |
| VANTAGE-Bench - Single Object Tracking | 30.38 | Single Object Tracking (%) | 31.2 |
| UGI Leaderboard | 28.23 | UGI Score | 25.9 |
| VANTAGE-Bench - Spatial | 58.28 | Spatial (%) | 12.5 |
| VANTAGE-Bench - Video QA | 63.85 | Video QA (%) | 12.5 |
| UGI - Writing | 15.19 | Writing Score | 6.4 |
| VANTAGE-Bench | 40.25 | Overall (%) | 6.2 |
| VANTAGE-Bench - Event Verification | 44.73 | Event Verification (%) | 6.2 |
| VANTAGE-Bench - Semantic | 54.29 | Semantic (%) | 6.2 |
| VANTAGE-Bench - Temporal | 18.03 | Temporal (%) | 6.2 |
Interactive version: theaggregate.ai/model?slug=qwen-3-vl-2b-instruct · How It Works · Data refreshed daily, snapshot 2026-08-05.