Long-form RewardBench: Evaluating Reward Models for Long-form Generation

ArXi:2603.12963v1 Announce Type: new The widespread adoption of reinforcement learning-based alignment highlights the growing importance of reward models. Various benchmarks have been built to evaluate reward models in various domains and scenarios. However, a significant gap remains in assessing reward models for long-form generation, despite its critical role in real-world applications. To bridge this, we