Global-MMLU-Lite: leaderboard
Lightweight multilingual MMLU benchmark from Cohere Labs and Kaggle for comparing knowledge performance across languages with reduced evaluation cost.
Metric: Accuracy (%). Source: www.kaggle.com. Source: artificialanalysis.ai. Saturation forecast: Estimated already saturated. 111 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 93.17 | #58 |
| 2 | Gemini 3 Pro (Preview) (High) | 92.22 | #66 (Gemini 3 Pro (Preview)) |
| 3 | Claude Opus 4.6 (Max) | 92.15 | #63 (Claude Opus 4.6) |
| 4 | Claude Opus 4.5 (Thinking) | 91.28 | #83 (Claude Opus 4.5) |
| 5 | GPT-5 (High) | 90.67 | #98 (GPT-5) |
| 6 | Claude Opus 4.5 (Non-reasoning) | 90.59 | #83 (Claude Opus 4.5) |
| 7 | GPT-5.1 (High) | 90.56 | #135 (GPT-5.1) |
| 8 | Claude Sonnet 4.6 (Max) | 90.48 | #90 (Claude Sonnet 4.6) |
| 9 | GPT-5 (Medium) | 90.39 | #98 (GPT-5) |
| 10 | Gemini 2.5 Pro | 90.31 | #152 |
| 11 | Qwen 3.5 397B A17B (Reasoning) | 90.04 | #146 (Qwen 3.5 397B A17B) |
| 12 | GPT-5.2 (Medium) | 89.76 | #110 (GPT-5.2) |
| 13 | GPT-5.1 Codex (High) | 89.7 | #159 (GPT-5.1 Codex) |
| 14 | Grok 4 | 89.49 | #181 |
| 15 | Grok 4.20 0309 (Reasoning) | 89.46 | #112 (Grok 4.20 0309) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=global-mmlu-lite · How It Works · Data refreshed daily, snapshot 2026-10-20.