ResearchBenchmarksPuzzle Solving on SimpleSokoban (final-step)Follow23.2Success RateActFocus-0.9285.33611.617.864May 14, 2026Evaluation ResultsMethodMethodLinksSuccess RateActFocusAlgorithm=GRPO, Model...Algorithm=GRPO, Model scale=3B, Evaluation step=2002026.0523.2UniformAlgorithm=GRPO, Model...Algorithm=GRPO, Model scale=3B, Evaluation step=2002026.050