Qwen VL Max (rolling alias)
Alibaba
Qwen-VL-Max is Alibaba Cloud Model Studio's mutable qwen-vl-max mainline API identifier rather than an immutable checkpoint. Alibaba has mapped it to multiple dated snapshots, including qwen-vl-max-2025-04-08 and qwen-vl-max-2025-08-13; this benchmark entry preserves results whose sources identify only the alias without establishing the served snapshot.
Benchmark scores
| Benchmark | Score |
|---|---|
| AI2D (OpenVLM) | 75.7 |
| CharXiv Reasoning | 24.7 |
| CharXiv-D | 41.5 |
| HallusionBench (OpenVLM) | 41.2 |
| MATH-Vision | 15.6 |
| MathVista (OpenVLM) | 43.6 |
| MM-Vet (OpenVLM) | 61.8 |
| MMMU Validation (OpenVLM) | 52.0 |
| MMStar (OpenVLM) | 49.5 |
| SnakeBench Average Apples | 9.2 |
| SnakeBench Best Apples | 22.0 |
| SnakeBench Rating | 27.6 |
| SnakeBench Total Apples | 129 |
| SnakeBench Win Rate | 58.3 |
| SpatialViz-Bench | 32.0 |
| Video-MME | 51.3 |