BiasBench — LLM Bias Leaderboard

BiasBench is an online evaluation platform that provides a systematic, standardized framework for assessing and comparing biases in large language models. Built on an extensive literature review, it organises LLM bias into four major dimensions — social, cultural, political and economic — covering 13 sub-dimensions, and reproduces 46 measurement methods from 21 peer-reviewed studies to score mainstream models with an output-based approach.

The results help developers locate and reduce bias, help educators teach critical thinking about AI outputs, and give policymakers evidence-based metrics for AI governance. Scores run from 0 to 100: a higher score means more pronounced bias, a lower score means the model is more neutral.