Multi-image Understanding on MMIU (test)
52.6AccuracyQwen2VL-7B
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2VL-7B2026.01 | 52.6 | |
| attention-masking strategybackbone=LLaVA-OV-7B, attention masking=true2026.01 | 45.5 | |
| LLaVA-OV-7B2026.01 | 45 | |
| InternVL2-Llama3-76B2026.01 | 44.2 | |
| Qwen2VL-2B2026.01 | 38.7 | |
| procedural data-generation strategybackbone=LLaVA-OV-0.5B, attention masking=false2026.01 | 37.2 | |
| InternVL2-8B2026.01 | 36.8 | |
| attention-masking strategybackbone=LLaVA-OV-0.5B, attention masking=true2026.01 | 36.3 | |
| LLaVA-OV-0.5B2026.01 | 34.2 | |
| LLaVA-v1.5-7B2026.01 | 19.2 | |
| InternVL2-2B2026.01 | 13.6 |