Visual Reasoning on AI2D
83.8AccuracyDualMindVLM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DualMindVLMSize=7B, Strategy=RL2025.11 | 83.8 | 104 | |
| ThinkLiteSize=7B, Strategy=RL2025.11 | 83.6 | 168 | |
| MM-EurekaSize=7B, Strategy=RL2025.11 | 83.5 | 207 | |
| OpenVLThinkerSize=7B, Strategy=SFT+RL2025.11 | 83.2 | 160 | |
| VL-RethinkerSize=7B, Strategy=RL2025.11 | 82.4 | 226 | |
| Qwen2.5-VLSize=7B, Strategy=-2025.11 | 80.8 | 145 |