Fine-grained Perception on VStarBench (Accuracy Breakdown)
89Overall AccuracySaGe-3B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SaGe-3BParam Size=3B2026.07 | 89 | 86.1 | 93.4 | |
| SaGe-7BParam Size=7B2026.07 | 89 | 85.2 | 94.7 | |
| DeepEyesParam Size=7B2026.07 | 85.6 | 87.3 | 84.5 | |
| ThymeParam Size=7B2026.07 | 82.2 | 83.5 | 80.3 | |
| LVRParam Size=7B2026.07 | 81.7 | 82.6 | 80.3 | |
| Qwen2.5-VLParam Size=32B2026.07 | 81.2 | 77.4 | 86.8 | |
| DyFoParam Size=7B2026.07 | 81.2 | 80 | 82.9 | |
| Pixel-ReasonerParam Size=7B2026.07 | 80.6 | 83.5 | 76.3 | |
| Vision-R1Param Size=7B2026.07 | 79.6 | 80 | 79 | |
| Gemini-2.5-ProParam Size=-2026.07 | 79.2 | — | — | |
| InternVL3Param Size=78B2026.07 | 76.4 | 75.7 | 77.6 | |
| Qwen2.5-VLParam Size=7B2026.07 | 76.4 | 77.4 | 75 | |
| Qwen2.5-VLParam Size=3B2026.07 | 75.4 | 80.9 | 67.1 | |
| SEALParam Size=7B2026.07 | 75.4 | 74.8 | 76.3 | |
| LLaVA-OVParam Size=72B2026.07 | 73.8 | 80.9 | 63.2 | |
| InternVL3Param Size=8B2026.07 | 72.3 | 73 | 71.1 | |
| LLaVA-OVParam Size=7B2026.07 | 70.7 | 73 | 60.5 | |
| GPT-4oParam Size=-2026.07 | 67.5 | 72.2 | 60.5 | |
| GPT-5-MiniParam Size=-2026.07 | 63.9 | — | — |