AA Global-MMLU-Lite: leaderboard
Metric: Accuracy (%). Source: artificialanalysis.ai. 112 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 93.17 |
| 2 | Gemini 3 Pro (Preview) (High) | 92.22 |
| 3 | Claude Opus 4.6 (Adaptive Reasoning, Max Effort) | 92.15 |
| 4 | Gemini 3 Flash (Preview) (Reasoning) | 91.44 |
| 5 | Claude Opus 4.5 (Thinking) | 91.28 |
| 6 | GPT-5 (High) | 90.67 |
| 7 | Claude Opus 4.5 (Non-reasoning) | 90.59 |
| 8 | GPT-5.1 (High) | 90.56 |
| 9 | Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort) | 90.48 |
| 10 | GPT-5 (Medium) | 90.39 |
| 11 | Gemini 2.5 Pro | 90.31 |
| 12 | Qwen 3.5 397B A17B (Reasoning) | 90.04 |
| 13 | GPT-5.2 (Medium) | 89.76 |
| 14 | GPT-5.1 Codex (High) | 89.7 |
| 15 | Grok 4 | 89.49 |
Interactive version: theaggregate.ai/benchmark?slug=aa-global-mmlu-lite · How It Works · Data refreshed daily, snapshot 2026-09-05.