ResearchBenchmarksGrid World Navigation on FrozenLake (final-step)Follow12.5Success RateUniform6.5728.1119.6511.189May 14, 2026Evaluation ResultsMethodMethodLinksSuccess RateUniformAlgorithm=GRPO, Model...Algorithm=GRPO, Model scale=3B, Evaluation step=2002026.0512.5ActFocusAlgorithm=GRPO, Model...Algorithm=GRPO, Model scale=3B, Evaluation step=2002026.056.8