MMLU-Pro

MMLU-Pro is the harder successor of MMLU: 14 subject areas, ten answer options instead of four and questions that require real reasoning instead of simply recognising trivia. This reduces random-guess luck and separates strong models more clearly.

MMLU-Pro measures broad factual and academic knowledge. In detail, it checks:

• knowledge combined with multi-step reasoning
• robustness against guessing (10 instead of 4 options)
• harder, more discriminative questions
• broad coverage of 14 academic areas

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

What is the MMLU-Pro benchmark?

MMLU-Pro is the harder successor of MMLU: 14 subject areas, ten answer options instead of four and questions that require real reasoning instead of simply recognising trivia. This reduces random-guess luck and separates strong models more clearly.

What does MMLU-Pro test?

MMLU-Pro mainly measures broad factual and academic knowledge. Concretely, it checks: knowledge combined with multi-step reasoning; robustness against guessing (10 instead of 4 options); harder, more discriminative questions.

How is the MMLU-Pro score calculated?

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.

Where do I get MMLU-Pro data and scores?

The official source is https://huggingface.co/datasets/TIGER-Lab/MMLU-Pro. The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/mmlu-pro/scores) or pull it directly from the Open LLM Radar API with any HTTP client.