group,benchmark_count,selected_average,best_checkpoint,best_average,selected_rank,selected_margin_vs_best Core Reasoning Tasks,3,0.719,checkpoints/step_900,0.726,2,-0.007 Language Understanding,4,0.754,checkpoints/step_900,0.756,2,-0.002 Generation Tasks,4,0.686,checkpoints/step_800,0.686,1,0.000 Specialized Capabilities,4,0.774,checkpoints/step_900,0.777,2,-0.002