Semifactual Credit-Augmented Policy Optimization

Benchmark Model Rank Results
mathematical-reasoning-on-omni-mathSCAPO (Qwen3-4B-Base)#4Accuracy: 39.42
mathematical-reasoning-on-olympiadbenchSCAPO (Qwen3-4B-Base)#20Accuracy: 56.82
mathematical-reasoning-on-minerva-mathSCAPO (Qwen3-4B-Base)#21Accuracy: 43.38
multiple-choice-question-answering-mcqa-on-gpqa-diamondSCAPO (Qwen3-4B-Base)#122Accuracy: 42.45