OneBench
One prompt is not enough: Instruction Sensitivity Undermines Embedding Model Evaluation | OneBench: AI Insights