List your company

Visual Reasoning on Maze, TSP, Sudoku, VSP, VSP-Super

92.1Average AccuracyEndoCoT
1.6225.1148.672.09Mar 12, 2026

Evaluation results

MethodLinks
EndoCoTTraining Setting=Task-Specific Training2026.0392.1
EndoCoTTraining Setting=Unified Training2026.0384.2
DiffThinkerTraining Setting=Task-Specific Training2026.0383.8
DiffThinkerTraining Setting=Unified Training2026.0377.1
Qwen3-VL-8BTraining Setting=Task-Specific Training, Optimization Method=SFT2026.0358.56
Qwen3-VL-8BTraining Setting=Task-Specific Training, Optimization Method=GRPO2026.0317.8
Qwen-Image-Edit-2511Training Setting=Zero-Shot2026.0311.7
Qwen3-VL-8BTraining Setting=Zero-Shot2026.0311.1
ChronoEditTraining Setting=Zero-Shot2026.038.8
ThinkGenTraining Setting=Zero-Shot2026.035.1