Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Global MMLU

Benchmarks

Task NameDataset NameSOTA ResultTrend
Factual knowledgeGlobal-MMLU-Lite
Seen Accuracy58.5
21
Knowledge-Based QAGlobal-MMLU Lite 5 Unseen Languages
MMLU Score71.9
20
Knowledge-Based QAGlobal-MMLU-Lite 8 Seen Languages
MMLU Score65.2
20
Multiple Choice Question AnsweringGlobal-MMLU Medical
Accuracy (ZH)89.1
17
Multilingual General KnowledgeGlobal MMLU Lite (subset of 18 languages)
Accuracy90.13
15
General KnowledgeGlobal MMLU Ukrainian (test)
Accuracy (%)67.03
14
Multilingual knowledge evaluationGlobal-MMLU Low
Accuracy52.31
12
Multilingual knowledge evaluationGlobal-MMLU High
Accuracy68.35
12
Multi-task Language UnderstandingGlobal MMLU-Lite Māori
Accuracy54.64
10
Multilingual Multiple-Choice ReasoningGlobal MMLU 42 languages 1.0 (test)
Average Accuracy54.8
6
Confidence EstimationGlobal-MMLU Japanese ja (test)
AUROC74
5
Confidence EstimationGlobal-MMLU Russian (test)
AUROC75
5
Confidence EstimationGlobal-MMLU Spanish es (test)
AUROC74
5
Language UnderstandingGlobal MMLU Overall
Accuracy59.2
5
Confidence EstimationGlobal-MMLU Japanese
AUROC0.72
2
Confidence EstimationGlobal-MMLU Russian
AUROC73
2
Confidence EstimationGlobal-MMLU Polish
AUROC77
2
Confidence EstimationGlobal-MMLU Spanish
AUROC78
2
Confidence EstimationGlobal-MMLU English
AUROC0.75
2
Confidence EstimationGlobal-MMLU French
AUROC0.76
2
Confidence EstimationGlobal-MMLU all languages average
AUROC0.5
2
General ReasoningGlobal MMLU 15 languages
Macro Accuracy54.77
2
Cross-lingual Reasoning and Factual KnowledgeGlobal MMLU (test)
Accuracy (RUS)23.46
2
Showing 23 of 23 rows