Skip to main content
BenchLM

Korean Massive Multitask Language Understanding (KMMLU)

We show this table for reference; we do not rank on it.

Data verified 36 confirmed releases in the last 30 daysFollow model changes

Evaluates Korean expert-level knowledge across 45 subjects. 20% of questions require Korean cultural context.

Benchmark score on KMMLU — September 29, 2026

We compile the KMMLU rows from provider self-reports. Kanana-2 3B Instruct leads the table at 43.3%, followed by Kanana-2 1.3B Instruct (42.8%). We do not use these results to rank models overall.

2 modelsKoreanKorean-language benchmarkRefreshingDisplay onlyUpdated September 29, 2026

Benchmark score table (2 models)

Score
1
Kanana-2 3B InstructKakao · Open weight
43.3%
2
Kanana-2 1.3B InstructKakao · Open weight
42.8%

About KMMLU

Year

2024

Tasks

35,030 questions

Format

Multiple choice questions

Difficulty

Elementary to professional level in Korean

Tests human-level understanding and reasoning in the Korean language across diverse subjects.

Freshness and provenance

Version

KMMLU 2024

Refresh cadence

Annual

Staleness state

Refreshing

Question availability

Public benchmark set

RefreshingDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does KMMLU measure?

Evaluates Korean expert-level knowledge across 45 subjects. 20% of questions require Korean cultural context.

Which model scores highest on KMMLU?

Kanana-2 3B Instruct by Kakao currently leads with a score of 43.3% on KMMLU.

How many models are evaluated on KMMLU?

2 AI models have been evaluated on KMMLU on BenchLM.

Last updated: September 29, 2026 · BenchLM version KMMLU 2024

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.