RewardBench

What is RewardBench?

RewardBench is a benchmark designed to evaluate the capabilities and safety of reward models (including those trained with Direct Preference Optimization, DPO).

Released

2025-06-02

Evaluates

指令跟随, Instruct, 其他, Other, 大语言模型, LLM, 指令遵循, Instruction Following, 不支持, Unsupported

Openness

restricted

Importer

OpenCompass Hub

Reported scores

0

Source provenance

• Original evidence https://hub.opencompass.org.cn/dataset-detail/RewardBench

RewardBench paper, code and dataset

• Paper https://arxiv.org/pdf/2506.01937
• Code https://github.com/allenai/reward-bench
• Dataset https://huggingface.co/datasets/allenai/blog-images
• Dataset https://huggingface.co/datasets/allenai/reward-bench-2
• Dataset https://huggingface.co/datasets/allenai/reward-bench-2-results

Which sources cite RewardBench?

• RewardBench

Related 指令跟随, Instruct, 其他, Other, 大语言模型, LLM, 指令遵循, Instruction Following, 不支持, Unsupported benchmarks

• A-Bench
• A-OKVQA
• AbsPyramid
• ActionAtlas
• Ada-LEval
• AECBench
• AFQMC
• AgentBoard