Fine-grained Multi-Image Understanding on VLM2-Bench
95.06Matching ScoreHuman-Level
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Human-LevelBase MLLM=-2026.04 | 95.06 | 98.11 | 94.23 | 96.02 | 91.29 | 92.87 | 97.08 | 91.17 | 94.48 | |
| MiCo-7BBase MLLM=Qwen2.5-VL2026.04 | 57.14 | 67.12 | 56.67 | 81.94 | 58 | 57.5 | 65 | 62 | 63.17 | |
| MM-Eureka-7BBase MLLM=Qwen2.5-VL2026.04 | 55.6 | 47.03 | 52.5 | 74.1 | 54 | 60 | 77.5 | 51 | 58.97 | |
| CGC-8BBase MLLM=Qwen3-VL2026.04 | 48.85 | 56.14 | 70.28 | 84.72 | 76 | 77.5 | 83 | 94 | 73.81 | |
| VLAA-Thinker-7BBase MLLM=Qwen2.5-VL2026.04 | 47.49 | 63.03 | 61.4 | 72.2 | 55 | 57.5 | 71 | 51 | 59.83 | |
| CGC-7BBase MLLM=Qwen2.5-VL2026.04 | 42.11 | 50.22 | 61.39 | 72.36 | 58.5 | 64.17 | 82 | 79 | 63.72 | |
| NoisyRollout-7BBase MLLM=Qwen2.5-VL2026.04 | 40.93 | 43.83 | 50.83 | 63.33 | 34.5 | 63.33 | 70.5 | 47 | 51.78 | |
| ThinkLite-VL-7BBase MLLM=Qwen2.5-VL2026.04 | 40.45 | 46.58 | 62.5 | 75.56 | 49.5 | 62.5 | 77.5 | 51 | 58.2 | |
| Qwen2.5-VL-7BBase MLLM=-2026.04 | 35.91 | 43.38 | 41.72 | 71.39 | 47.5 | 59.76 | 80 | 69 | 56.08 | |
| InternVL2.5-26BBase MLLM=-2026.04 | 30.5 | 30.59 | 51.48 | 43.33 | 52.5 | 59.67 | 59.5 | 61.25 | 48.6 | |
| Qwen3-VL-8BBase MLLM=-2026.04 | 27.03 | 48.86 | 66.39 | 80 | 70.5 | 73.33 | 80 | 88 | 66.76 | |
| InternVL3-8BBase MLLM=-2026.04 | 23.94 | 29.68 | 64.17 | 61.67 | 39.5 | 55 | 63 | 55 | 49 | |
| InternVL2.5-8BBase MLLM=-2026.04 | 21.24 | 26.53 | 55.23 | 53.33 | 46.5 | 60 | 51.5 | 52 | 45.79 | |
| LLaVA-Video-7BBase MLLM=-2026.04 | 18.53 | 12.79 | 62.47 | 54.72 | 28.5 | 66.91 | 62 | 25 | 41.37 | |
| Qwen2-VL-7BBase MLLM=-2026.04 | 18.07 | 19.18 | 61.84 | 68.08 | 37.5 | 67.92 | 72 | 47 | 48.95 | |
| mPLUG-Owl2-7BBase MLLM=-2026.04 | 17.37 | 18.26 | 62.97 | 49.17 | 31 | 58.06 | 63 | 29 | 41.1 | |
| LLaVA-OV-7BBase MLLM=-2026.04 | 16.6 | 13.7 | 56.17 | 47.22 | 27.5 | 46.67 | 62 | 37 | 38.36 | |
| LongVQA-7BBase MLLM=-2026.04 | 14.29 | 12.98 | 49.47 | 46.53 | 29 | 41.56 | 58 | 25 | 34.6 |