Global MMLU Lite (Kaggle)

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

Kaggle reports an average across 16 language dimensions of Global MMLU Lite, with separate culturally sensitive and culturally agnostic results.

Evaluation results published by Cohere Labs. How Kaggle Benchmarks reruns tests.

Kaggle average on Global MMLU Lite (Kaggle) v1 — Source retrieved 2026-10-20T15:19:04.746Z

We mirror the published kaggle average view for Global MMLU Lite (Kaggle) v1. Claude Opus 5 has the highest reported value at 95.7%. These are published source configurations. Compare the task version and disclosed settings before treating a score difference as a model difference. We do not use these results to rank models overall.

Claude Opus 5

95.7%

Gemini 3.5 Flash

95.4%

Gemini 4 Argon

95.4%

45 modelsMultilingualCurrentDisplay onlyUpdated Source retrieved 2026-10-20T15:19:04.746Z

Kaggle average table (45 models)

Kaggle average results for Global MMLU Lite (Kaggle)
Rank	Model / configuration	Score	Parameters (B)	Open / closed
1	Claude Opus 5Creator not reported	95.7%	Not reported	Unknown
2	Gemini 3.5 FlashCreator not reported	95.4%	Not reported	Unknown
3	Gemini 4 ArgonCreator not reported	95.4%	Not reported	Unknown
4	Claude Opus 5.5Creator not reported	95.4%	Not reported	Unknown
5	Gemini 3.8 FlashCreator not reported	95.3%	Not reported	Unknown
6	Gemini 3.7 FlashCreator not reported	94.9%	Not reported	Unknown
7	Claude Opus 4.8Creator not reported	94.9%	Not reported	Unknown
8	Gemini 3 Pro PreviewCreator not reported	94.5%	Not reported	Unknown
9	Claude Opus 4.1Creator not reported	94.3%	Not reported	Unknown
10	Gemini 3.6 FlashCreator not reported	94.3%	Not reported	Unknown
11	GPT-5.6 SolCreator not reported	94.1%	Not reported	Unknown
12	GPT-6 SolCreator not reported	94.1%	Not reported	Unknown
13	Grok 4.5Creator not reported	93.3%	Not reported	Unknown
14	Gemini 2.5 ProCreator not reported	93.2%	Not reported	Unknown
15	GPT-5.6 TerraCreator not reported	92.7%	Not reported	Unknown
16	Claude Sonnet 5.5Creator not reported	92.6%	Not reported	Unknown
17	GPT-6 AstraCreator not reported	92.6%	Not reported	Unknown
18	GPT-6 LunaCreator not reported	92.0%	Not reported	Unknown
19	Gemini 2.5 FlashCreator not reported	91.5%	Not reported	Unknown
20	Gemini 2.5 Flash PreviewCreator not reported	90.9%	Not reported	Unknown
21	Gemma 4 31BCreator not reported	90.7%	Not reported	Unknown
22	Claude Sonnet 4Creator not reported	90.6%	Not reported	Unknown
23	Gemini 3.5 Flash-LiteCreator not reported	89.5%	Not reported	Unknown
24	GPT-5Creator not reported	89.0%	Not reported	Unknown
25	Grok 4Creator not reported	88.8%	Not reported	Unknown
26	Qwen 3 235B A22B Instruct 2506Creator not reported	88.0%	Not reported	Unknown
27	GPT-4.1Creator not reported	87.5%	Not reported	Unknown
28	Claude Haiku 5.5Creator not reported	87.3%	Not reported	Unknown
29	o4 miniCreator not reported	87.0%	Not reported	Unknown
30	Gemma 4 26B A4BCreator not reported	85.7%	Not reported	Unknown
31	Command A Creator not reported	83.9%	Not reported	Unknown
32	GPT-5.6 LunaCreator not reported	83.1%	Not reported	Unknown
33	Claude 3.7 SonnetCreator not reported	80.8%	Not reported	Unknown
34	deepseek-v3.1Creator not reported	80.4%	Not reported	Unknown
35	GPT-6.1 SolCreator not reported	78.9%	Not reported	Unknown
36	Mistral Small 3.1Creator not reported	78.5%	Not reported	Unknown
37	o3 miniCreator not reported	78.0%	Not reported	Unknown
38	Gemma 3 27BCreator not reported	76.3%	Not reported	Unknown
39	Granite 4.0 SmallCreator not reported	75.0%	Not reported	Unknown
40	Aya Expanse 32BCreator not reported	73.5%	Not reported	Unknown
41	DeepSeek-R1Creator not reported	67.4%	Not reported	Unknown
42	Grok 3 MiniCreator not reported	67.3%	Not reported	Unknown
43	Gemma 3 4BCreator not reported	65.1%	Not reported	Unknown
44	Claude 3.5 HaikuCreator not reported	61.1%	Not reported	Unknown
45	Mistral Medium 3Creator not reported	55.1%	Not reported	Unknown

How to read this leaderboard

Compare multilingual multiple-choice knowledge within the Kaggle protocol. Read the culturally sensitive and culturally agnostic scores beside the average; the dataset language count is not the denominator of this leaderboard average.

Operator receipt: 45 sourced rows are currently displayable on this page; the leading published row is Claude Opus 5 at 95.7%.

Honest limit: This Kaggle protocol is separate from Artificial Analysis Global-MMLU-Lite and the full Global-MMLU dataset. The source response includes task version 1 for Average and task version 2 for language and cultural-sensitivity metrics. Unspecified model settings remain unknown. These results do not enter weighted rankings.

Kaggle leaderboard and Apache 2.0 declaration Published leaderboard download Cohere Labs Global-MMLU-Lite dataset card Global-MMLU paper Apache 2.0 license copy

How we show Global MMLU Lite (Kaggle)

The source dataset has 400 questions per language, split evenly between culturally sensitive and culturally agnostic samples, using human translations or post-edits. Its current dataset card covers 23 languages; this Kaggle v1 board instead labels its published Average as an average of 16 language dimensions. We preserve that published average and the individual source metrics without recomputing a broader score.

This Kaggle protocol is separate from Artificial Analysis Global-MMLU-Lite and the full Global-MMLU dataset. The source response includes task version 1 for Average and task version 2 for language and cultural-sensitivity metrics. Unspecified model settings remain unknown. These results do not enter weighted rankings.

Snapshot

45 source rowsKaggle v1Apache 2.0Display only

Kaggle leaderboard and Apache 2.0 declaration Cohere Labs Global-MMLU-Lite dataset card Global-MMLU paper

These are published source configurations. Compare the task version and disclosed settings before treating a score difference as a model difference. The table retains each source configuration and its reported value.

45 models are shown for Global MMLU Lite (Kaggle). The benchmark falls in the Multilingual category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Global MMLU Lite (Kaggle) is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About Global MMLU Lite (Kaggle)

Year

2026

Tasks

400 test questions per language; source average of 16 language dimensions

Format

Source-published percentage score

Difficulty

Multilingual multiple-choice knowledge

Public benchmark source

Freshness and provenance

Version

Global MMLU Lite (Kaggle) 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay 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 Global MMLU Lite (Kaggle) measure?

Kaggle reports an average across 16 language dimensions of Global MMLU Lite, with separate culturally sensitive and culturally agnostic results.

Which source row has the highest reported Global MMLU Lite (Kaggle) value?

Claude Opus 5 has the highest reported Global MMLU Lite (Kaggle) value at 95.7%. These are published source configurations. Compare the task version and disclosed settings before treating a score difference as a model difference. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on Global MMLU Lite (Kaggle)?

The Source retrieved 2026-10-20T15:19:04.746Z snapshot contains 45 AI models.

Last updated: Source retrieved 2026-10-20T15:19:04.746Z · mirrored from the public benchmark leaderboard

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 6,100+ subscribers.

One email each week. Unsubscribe anytime.