Multimodal Reasoning on MMStar (ACC%, S, I)
67.1Accuracy (ACC%)Qwen3-VL-30B+VRGA
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-VL-30B+VRGABackbone=Qwen3-VL-30B, Enhancement=VRGA2026.03 | 67.1 | 0.522 | 0.543 | |
| Qwen3-VL-30BBackbone=Qwen3-VL-30B2026.03 | 66.1 | 0.521 | 0.542 | |
| Qwen2-VL-7BBackbone=Qwen2-VL-7B2026.03 | 52.37 | 0.419 | 0.499 | |
| Qwen2-VL-7B+VRGABackbone=Qwen2-VL-7B, Enhancement=VRGA2026.03 | 51.07 | 0.405 | 0.513 | |
| Qwen2.5-VL-3B+VRGABackbone=Qwen2.5-VL-3B, Enhancement=VRGA2026.03 | 50.93 | 0.441 | 0.383 | |
| Qwen2-VL-7B+CCOTBackbone=Qwen2-VL-7B, Enhancement=CCOT2026.03 | 50.06 | — | — | |
| Qwen2.5-VL-3BBackbone=Qwen2.5-VL-3B2026.03 | 49.8 | 0.436 | 0.382 | |
| Qwen2.5-VL-7B+VRGABackbone=Qwen2.5-VL-7B, Enhancement=VRGA2026.03 | 48.73 | 0.388 | 0.569 | |
| Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B2026.03 | 48.53 | 0.389 | 0.557 | |
| Qwen2.5-VL-3B+CCOTBackbone=Qwen2.5-VL-3B, Enhancement=CCOT2026.03 | 46.27 | — | — | |
| Qwen2-VL-7B+ICOTBackbone=Qwen2-VL-7B, Enhancement=ICOT2026.03 | 42.27 | 0.358 | 0.523 |