# What is KMMLU? The Korean Large Language Model Benchmark Explained

> KMMLU measures whether a model can reason across Korean-native exams and professional subject matter, rather than merely translating competence from English benchmarks.

- Canonical page: https://benchlm.ai/guides/kmmlu-explained
- **Last verified:** September 22, 2026
- **Questions:** 35,030 multiple-choice items
- **Coverage:** 45 localized subject areas
- **Korean-specific content:** More than 20% of the benchmark requires Korean historical, legal, geographic, or cultural knowledge.

## Why translated English benchmarks are not enough

Translated sets can preserve a math problem while losing the local context that matters in law, education, government, history, and culture. KMMLU uses Korean-native expert and exam material, making it a stronger signal for real Korean-language deployments.

## How KMMLU is structured

| Supercategory | Examples |
|---|---|
| STEM | Mathematics, physics, chemistry, and software engineering |
| Applied science | Medicine, telecommunications, and civil engineering |
| HUMSS | Korean history, constitutional law, sociology, and humanities |
| Other professional fields | Patent law, real estate, and specialist domains |

## How to interpret the benchmark family

- KMMLU is the broad Korean-native baseline.
- KMMLU-Hard and KMMLU-Pro remove easier rows and better separate frontier systems.
- CLIcK and KoBALT add culturally grounded and adjacent Korean-language evidence.
- Global models may lead on raw reasoning, while Korean-market specialists can separate on legal, historical, linguistic, and cultural tasks.

## Continue with live evidence

- [Korean benchmarks leaderboard](/leaderboards/korean-benchmarks)
- [Korean-market LLM leaderboard](/leaderboards/korean-llm)
- [KMMLU benchmark profile](/benchmarks/kmmlu)
- [KMMLU-Hard benchmark profile](/benchmarks/kmmluhard)
- [KMMLU-Pro benchmark profile](/benchmarks/kmmlupro)

## Primary source

- [KMMLU: Measuring Massive Multitask Language Understanding in Korean](https://arxiv.org/abs/2402.11548)