Qwen 3 VL 32B 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 1663 ± 4, rank #283 of 1656 rated models, from 923 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Appear2Meaning | 2.9 | Exact-match accuracy (%, times 100; every predicted metadata | 100 |
| AutoGUI-v2 - Element Grounding (Appearance) | 80.7 | Element grounding by appearance description on the AutoGUI-v | 100 |
| AutoGUI-v2 - Element Grounding (Functionality) | 71.1 | Element grounding by functionality (what an interaction does | 100 |
| AutoGUI-v2 - Element Grounding (Intent) | 81.9 | Element grounding by action intent on the AutoGUI-v2 element | 100 |
| CulMind - Seals | 53.1 | Seals subdomain score: macro-average of the task scores (0-1 | 100 |
| CulMind-R | 64.9 | Final-answer score (0-100) on the 24-task CulMind-R reasonin | 100 |
| EgoPoint-Bench - Simulation Adversarial Resilience | 60.23 | Accuracy (%) on the Adversarial Resilience questions among t | 100 |
| EgoPoint-Bench - Simulation OCR | 79.39 | Accuracy (%) on the OCR questions among the 1,929 held-out s | 100 |
| Enginuity - Parts Recall (Direct) | 86.5 | Recall@all (%): share of ground-truth parts recovered, match | 100 |
| FIND | 77.8 | Zero shot en (self-reported) | 100 |
| HoloCount - Null-Target Prompting | 98 | Accuracy (%; exact-match accuracy of the single integer coun | 100 |
| KnowHal - Entity (Negative) | 52.9 | Accuracy (%; entity-dimension negative (false-premise) quest | 100 |
Interactive version: theaggregate.ai/model?slug=qwen-3-vl-32b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.