Psychometric Audit Reveals MMLU Measures Retrieval Rather Than Reasoning
September 10, 2026
A psychometric analysis of 1,000 open-weights models shows MMLU scores primarily reflect factual retrieval capacity rather than reasoning. The study demonstrates that MMLU conflates separable constructs and that difficulty is driven by text-extractable structural complexity.
HOW THIS AFFECTS YOU
●
researcherYou should avoid using aggregate MMLU scores as a sole proxy for model reasoning capabilities.