Qwen 3 VL 30B A3B (Thinking): benchmark results
Alibaba Qwen 3 VL 30B A3B vision-language model evaluated with thinking enabled. Provider: Alibaba. Released 2025-10-04. Access: Open.
Aggregate Rating 1573 ± 6, rank #784 of 2168 rated models, from 477 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EarthSpatialBench (CoT) - Angle Estimation | 54.99 | Mean absolute error in degrees of the estimated azimuth betw | 100 |
| EarthSpatialBench (Direct) - Direction Binary Classification | 71 | Macro-averaged F1 (%) on EarthSpatialBench yes/no direction | 100 |
| HAERAE-Vision - Mathematics | 68.69 | Checklist Score (%; original under-specified queries, GPT-5- | 100 |
| VIA-Bench (Pattern Match) - Motion Illusions | 96.28 | Accuracy (%) on the motion illusions questions, answer lette | 100 |
| VideoVIBE - Layout Integrity | 82.8 | Failure-diagnosis accuracy (%; questions of the Layout Integ | 100 |
| VideoVIBE - Motion Quality | 84.56 | Failure-diagnosis accuracy (%; questions of the Motion Quali | 100 |
| VIA-Bench (LLM Judge) - Motion Illusions | 96.28 | Accuracy (%) on the motion illusions questions, answer lette | 95.5 |
| GeoAux-Bench (Core) - Olympiad Junior | 93.24 | Accuracy (%) on the 207 junior olympiad questions of GeoAux- | 95 |
| GeoAux-Bench (Core) - Olympiad Senior | 83.97 | Accuracy (%) on the 237 senior olympiad questions of GeoAux- | 95 |
| OMIBench - Chemistry | 32.26 | GPTScore (%, binarized LLM-judge equivalence of the final an | 94.4 |
| OMIBench - Mathematics | 59.3 | GPTScore (%, binarized LLM-judge equivalence of the final an | 94.4 |
| HAERAE-Vision - Health/Medical | 56.38 | Checklist Score (%; original under-specified queries, GPT-5- | 93.2 |
Interactive version: theaggregate.ai/model?slug=qwen-3-vl-30b-a3b-thinking · How It Works · Data refreshed daily, snapshot 2026-10-20.