Global-MMLU-Lite Benchmark Leaderboard

Multilingual & MultimodalLive

Leaderboard

Top 14 of 14 models we hold a score for.

1Gemini 2.5 Pro Preview 06-05Google

89.2%2Gemini 2.5 ProGoogle

88.6%3Gemini 2.5 FlashGoogle

88.4%4Gemini 2.5 Flash-LiteGoogle

81.1%5Gemini 2.0 Flash-LiteGoogle

78.2%6Gemma 3 27BGoogle

75.1%7Gemma 3 12BGoogle

69.5%8Gemini DiffusionGoogle

69.1%9Gemma 3n E4B Instructed LiteRT PreviewGoogle

64.5%10Gemma 3n E4B InstructedGoogle

64.5%11Gemma 3n E2B Instructed LiteRTPreviewGoogle

59.0%12Gemma 3n E2B InstructedGoogle

59.0%13Gemma 3 4BGoogle

54.5%14Gemma 3 1BGoogle

34.2%

Source

Plain explanation

What it measures, how to read the number, and what to watch out for.

A trimmed, translated MMLU built to check whether a model’s knowledge survives the trip out of English, across a spread of languages and cultural contexts. Higher is better, but the number worth watching is not the average — it is the spread between a model’s best and worst languages, which is where multilingual weakness actually shows up. Translated benchmarks inherit the quality of their translations, and some questions are culturally specific in ways that make the "correct" answer arguable outside the context they came from. Coverage is deliberately light, so treat it as a screening test rather than a serious multilingual evaluation.