Research
RewardBench 2: Advancing Reward Model Evaluation
Reward models are used throughout the post-training of language models to capture nuanced signals from preference data and provide a training target for optimization across instruction following, reasoning, safety, and more domains.
