Qwen 3 VL 8B 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 2025-10-15. Access: Open.
Aggregate Rating 1606 ± 3, rank #420 of 1656 rated models, from 1559 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| BoxComm | 23.7 | Content consistency accuracy (%) over all requested categori | 100 |
| BoxComm (Punch Events) | 23.4 | Content consistency accuracy (%) over all requested categori | 100 |
| BoxComm (Punch Events) - Contextual | 23.3 | Content consistency accuracy (%) on the sentences whose requ | 100 |
| BoxComm (Punch Events) - Tactical | 28 | Content consistency accuracy (%) on the sentences whose requ | 100 |
| BoxComm - Contextual | 18.3 | Content consistency accuracy (%) on the sentences whose requ | 100 |
| BoxComm - Tactical | 32 | Content consistency accuracy (%) on the sentences whose requ | 100 |
| CVSBench - Viewpoint Arrow | 42.3 | Accuracy (%) on View-Arrow: choose the camera-pose arrow on | 100 |
| CoCoBench - Task Allocation | 95 | Task success rate (%; task-allocation instances, dividing in | 100 |
| CulMind - Ancient Scripts | 19.1 | Ancient Scripts subdomain score: macro-average of the task s | 100 |
| CulMind - Calligraphy | 65.1 | Calligraphy subdomain score: macro-average of the task score | 100 |
| DatBench - General VQA | 63.7 | Accuracy (%) | 100 |
| DefectBench (Building Facades) - Crack Counting Error | 0.62 | Mean absolute error of the model's count of crack instances | 100 |
Interactive version: theaggregate.ai/model?slug=qwen-3-vl-8b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.