List your company

Interactive game puzzle solving on Sokoban 6x6 (test)

86.98Average Task ScoreGraphGPO
8.688829.014449.3469.6656May 26, 2026

Evaluation results

MethodLinks
GraphGPOType=RL Training, Base Model=Qwen2.5-VL-3B-Instruct2026.0586.9886.98
GiGPOType=RL Training, Base Model=Qwen2.5-VL-3B-Instruct2026.0576.9276.92
GRPOType=RL Training, Base Model=Qwen2.5-VL-3B-Instruct2026.0567.167.1
Qwen2.5-VLType=Prompting, Base Model=Qwen2.5-VL-3B-Instruct2026.0511.711.7