GreekMMLU is a native-sourced benchmark for evaluating massive multitask language understanding in Greek, built from authentic Greek exam-style multiple-choice questions (MCQ) rather than machine-translated English benchmarks.

21,805 questions across 45 subjects
4 high-level groups: STEM, Humanities, Social Sciences, Other
Difficulty/education levels spanning Primary → Secondary → University → Professional (+ an N/A bucket)
Public vs. private split for contamination-resistant evaluation: 16,857 public/4,948 private (leaderboard)

Structure

Configs: Each subject is a separate configuration (subset).
Splits:
- test: Public test set (High volume)
- dev: Dev/Few-shot set (5-shot)

Usage

from datasets import load_dataset

# Load specific subject 
ds = load_dataset("dascim/GreekMMLU", "Agriculture_Professional")

# Load all data combined
ds = load_dataset("dascim/GreekMMLU", "All")

What makes GreekMMLU different?

Most “Greek MMLU” style evaluations rely on machine translation from English. Instead, GreekMMLU uses original Greek content sourced or authored from real educational/professional assessments, aiming to preserve:

Greek morphology and punctuation
Greek-specific cultural/institutional knowledge (e.g., Greek History, Greek Traditions)
Realistic exam difficulty calibration

Task format

Multiple choice, 2–4 options, exactly one correct. The harness prompt uses Greek option labels (Α, Β, Γ, Δ) for a fully native evaluation setup.

Subjects (high-level)

GreekMMLU includes 45 subjects, grouped into:

Humanities (e.g., Art; Greek History; Greek Literature; Greek Mythology; Law; World Religions)
STEM (e.g., Mathematics; Physics; Computer Science; Electrical Engineering; Medicine)
Social Sciences (e.g., Economics; Education; Government & Politics; Modern Greek Language; Accounting)
Other (e.g., Driving Rules; General Knowledge; Maritime Safety and Rescue Operations) See the paper for the full taxonomy and educational-level breakdown.

Citation

If you use GreekMMLU in your work, please cite the paper:

@article{zhang2026greekmmlu,
  title={GreekMMLU: A Native-Sourced Multitask Benchmark for Evaluating Language Models in Greek},
  author={Zhang, Yang and Konomi, Mersin and Xypolopoulos, Christos and Divriotis, Konstantinos and Skianis, Konstantinos and Nikolentzos, Giannis and Stamou, Giorgos and Shang, Guokan and Vazirgiannis, Michalis},
  journal={arXiv preprint arXiv:2602.05150},
  year={2026}
}