Visual Reasoning on Maze
100Accuracy (Scale 8)DiffThinker
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DiffThinkerTraining Setting=Task-Specific Training2026.03 | 100 | 100 | 65 | |
| EndoCoTTraining Setting=Task-Specific Training2026.03 | 100 | 100 | 90 | |
| DiffThinkerTraining Setting=Unified Training2026.03 | 98 | 99 | 66 | |
| EndoCoTTraining Setting=Unified Training2026.03 | 97 | 98 | 52 | |
| Qwen3-VL-8BTraining Setting=Task-Specific Training, Optimization Method=SFT2026.03 | 53 | 37 | 0 | |
| ChronoEditTraining Setting=Zero-Shot2026.03 | 1 | 1 | 26 | |
| Qwen3-VL-8BTraining Setting=Zero-Shot2026.03 | 1 | 0 | 0 | |
| ThinkGenTraining Setting=Zero-Shot2026.03 | 0 | 0 | 0 | |
| Qwen-Image-Edit-2511Training Setting=Zero-Shot2026.03 | 0 | 0 | 0 | |
| Qwen3-VL-8BTraining Setting=Task-Specific Training, Optimization Method=GRPO2026.03 | 0 | 0 | 0 |