List your company

Reasoning on Sokoban

99.1Success RateBFS
73.82880.38986.9593.511May 29, 2026

Evaluation results

MethodLinks
BFSTraining=RL2026.0599.164.08
With parent pointerTraining=GRPO + generation constraint2026.0598.954.7
BFSTraining=pretrained2026.0594.973.21
With parent pointerTraining=GRPO2026.0589.652.82
No parent pointerTraining=GRPO2026.0585.963.54
With parent pointerTraining=SFT2026.0585.673.26
No parent pointerTraining=SFT2026.0574.868.24