Multimodal Perception and Reasoning on MME-RealWorld Lite (Detailed Metrics)
57.4Overall ScoreP2R-8B
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| P2R-8BBackbone=Qwen3-VL-Instruct, Model Size=8B, Method=P2R2026.07 | 57.4 | 92.4 | 58.7 | 89 | 44.5 | 46.9 | 81 | 81 | 46.7 | 39 | |
| P2R-4BBackbone=Qwen3-VL-Instruct, Model Size=4B, Method=P2R2026.07 | 54.8 | 88.8 | 50 | 86 | 46.1 | 39.1 | 79 | 78 | 46.7 | 39.5 | |
| DeepEyes-7BBackbone=DeepEyes, Model Size=7B2026.07 | 53.2 | 90 | 52.7 | 89 | 43.3 | 33.4 | 76 | 69 | 44 | 35 | |
| Qwen3-VL-Instruct-32BBackbone=Qwen3-VL-Instruct, Model Size=32B, Method=Base2026.07 | 52.3 | 86.8 | 50 | 89 | 38.6 | 36 | 82 | 88 | 44 | 34.5 | |
| P2R-2BBackbone=Qwen3-VL-Instruct, Model Size=2B, Method=P2R2026.07 | 51.3 | 88.4 | 43.3 | 78 | 39.8 | 40.6 | 76 | 68 | 44 | 35.5 | |
| Qwen3-VL-Instruct-8BBackbone=Qwen3-VL-Instruct, Model Size=8B, Method=Base2026.07 | 50.4 | 88.8 | 49.3 | 85 | 36.4 | 33.1 | 81 | 79 | 38.7 | 34.3 | |
| PixelReasoner-7BBackbone=PixelReasoner, Model Size=7B2026.07 | 49.7 | 89.6 | 52 | 86 | 38.9 | 30.9 | 71 | 72 | 46 | 32.5 | |
| Qwen3-VL-Instruct-4BBackbone=Qwen3-VL-Instruct, Model Size=4B, Method=Base2026.07 | 47.7 | 88 | 38 | 80 | 34.8 | 33.7 | 74 | 77 | 34.7 | 31.8 | |
| Qwen3-VL-Instruct-2BBackbone=Qwen3-VL-Instruct, Model Size=2B, Method=Base2026.07 | 47.3 | 85.2 | 38.7 | 74 | 32.9 | 39.1 | 69 | 67 | 40.7 | 30.8 |