DeepSeek-GRM models were able to outperform existing methods, achieving a competitive performance with strong public reward models.