#llm-benchmarks

LLM Harness Sensitivity: How Benchmark Choices Skew AI Model Rankings

A new study shows that small changes in evaluation harness configuration can flip leaderboard positions, raising questions about the reliability of current LLM benchmarks.