README.md

metadata

title: PRISM Benchmark
emoji: 📊
colorFrom: purple
colorTo: indigo
sdk: docker
pinned: false

PRISM: A Multi-Dimensional Benchmark for Evaluating LLM Peer Reviewers

This is a Hugging Face Space hosting the official project page for PRISM, a comprehensive benchmark for evaluating Large Language Models (LLMs) as peer reviewers.

About PRISM

PRISM is a multi-dimensional benchmark designed to evaluate how well LLMs can perform peer review tasks. The benchmark covers multiple aspects of academic review including:

• Constructiveness: Evaluating the quality and helpfulness of feedback
• Novelty Assessment: Identifying novel contributions in research papers
• Flaw Detection: Identifying weaknesses and issues in academic work
• Evidence-Based Reasoning: Assessing how well reviewers ground their feedback

Links

• Project Page: https://prism-benchmark.github.io
• Demo: https://prism-benchmark.github.io/demo/
• ArXiv Paper: https://arxiv.org/abs/2605.26730
• Hugging Face Paper: https://huggingface.co/papers/2605.26730
• Full Code Repository: https://github.com/prism-benchmark/prism
• Website Repository: https://github.com/prism-benchmark/prism-benchmark.github.io