Multimodal Understanding on Composite VLM Benchmark Suite
74.5Overall Average ScoreSAIL-VL
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SAIL-VLModel Size=8B2025.01 | 74.5 | 68.3 | 79.8 | 83.3 | 68.7 | |
| InternVL2.5-MPOModel Size=8B2025.01 | 74.3 | 71.2 | 76.3 | 83.2 | 69.7 | |
| Qwen2-VLModel Size=8B2025.01 | 73 | 68.5 | 79.6 | 71 | 67.5 | |
| DeepSeekVL-2Model Size=8B2025.01 | 72.7 | 66.8 | 79 | 79 | 65.3 | |
| SAIL-VLModel Size=2B2025.01 | 69.1 | 60.4 | 75.9 | 79 | 66.2 | |
| InternVL2.5-MPOModel Size=2B2025.01 | 67.7 | 63.1 | 71.1 | 75.3 | 64.5 | |
| DeepSeekVL-2Model Size=2B2025.01 | 67 | 59.4 | 74.4 | 71.3 | 63.6 | |
| Qwen2-VLModel Size=2B2025.01 | 64.4 | 58.3 | 72.5 | 59 | 62.9 |