ResearchBenchmarksTest-time Alignment on HH-RLHFFollow86Win Rate vs LlamaCAST85.27285.46185.6585.839Apr 26, 2026Evaluation ResultsMethodMethodLinksWin Rate vs LlamaWin Rate vs DeepSeekWin Rate vs GPTCASTBase Model=Hermes3-8B,...Base Model=Hermes3-8B, Alignment Stage=Test-time2026.048679.974.7Pref-CTRLBase Model=Hermes3-8B,...Base Model=Hermes3-8B, Alignment Stage=Test-time2026.0485.784.373.6Best-of-NBase Model=Hermes3-8B,...Base Model=Hermes3-8B, Alignment Stage=Test-time2026.0485.378.972.1