Visual Reasoning on Maze, TSP, Sudoku, VSP, VSP-Super
92.1Average AccuracyEndoCoT
Evaluation Results
| Method | Links | |
|---|---|---|
| EndoCoTTraining Setting=Task-Specific Training2026.03 | 92.1 | |
| EndoCoTTraining Setting=Unified Training2026.03 | 84.2 | |
| DiffThinkerTraining Setting=Task-Specific Training2026.03 | 83.8 | |
| DiffThinkerTraining Setting=Unified Training2026.03 | 77.1 | |
| Qwen3-VL-8BTraining Setting=Task-Specific Training, Optimization Method=SFT2026.03 | 58.56 | |
| Qwen3-VL-8BTraining Setting=Task-Specific Training, Optimization Method=GRPO2026.03 | 17.8 | |
| Qwen-Image-Edit-2511Training Setting=Zero-Shot2026.03 | 11.7 | |
| Qwen3-VL-8BTraining Setting=Zero-Shot2026.03 | 11.1 | |
| ChronoEditTraining Setting=Zero-Shot2026.03 | 8.8 | |
| ThinkGenTraining Setting=Zero-Shot2026.03 | 5.1 |