#reward model

'ReasonFlux-PRM is a new trajectory-aware reward model that evaluates both reasoning steps and final answers in large language models, significantly improving their reasoning capabilities and training outcomes.'