Metric & Benchmark
aka Massive Multitask Language Understanding
MMLU
57-subject multiple-choice benchmark spanning STEM, humanities and professional exams.
Definition
MMLU tests an LLM's broad knowledge across 57 academic and professional subjects, from elementary math to law and medicine. It is the standard general-knowledge benchmark; frontier models exceed 88% and use MMLU-Pro as the harder follow-up.
Common use cases
- Model comparison
- General knowledge eval