Qwen2.5-VL-7B-Instruct
Alibaba · 2025-01-26 · 8.3B parameters
Qwen2.5-VL-7B-Instruct is Alibaba's 7B-class instruction-tuned vision-language checkpoint from the Qwen2.5-VL series, released January 26, 2025 under Apache 2.0. Its complete checkpoint contains exactly 8,292,166,656 parameters and supports dynamic-resolution images, long-video understanding, OCR, document and chart analysis, UI grounding, stable structured outputs, and visual-agent workflows for computers and phones.
Benchmark scores
| Benchmark | Score |
|---|---|
| Capability | 121.5 |
| CharXiv Reasoning | 42.5 |
| CharXiv-D | 73.9 |
| MATH-Vision | 25.1 |
| ScreenSpot-Pro | 26.8 |
| SnakeBench Average Apples | 0.2 |
| SnakeBench Best Apples | 1.0 |
| SnakeBench Rating | 8.3 |
| SnakeBench Total Apples | 4.0 |
| SnakeBench Win Rate | 11.1 |