Mobile-MMLU: leaderboard
Mobile-MMLU benchmark testing factual knowledge across subjects with prompts tailored to small on-device LLMs (overall accuracy).
Metric: Overall Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 2 9B (IT) | 75 | #788 |
| 2 | Qwen 2.5 7B Instruct | 74.9 | #862 |
| 3 | Yi-1.5-9B Chat | 72.7 | #1031 |
| 4 | Ministral-8B-Instruct-2410 | 71.5 | #1094 |
| 5 | Qwen 2.5 3B Instruct | 68.1 | #1162 |
| 6 | Llama 3.1 8B Instruct | 66.9 | #1035 |
| 7 | InternLM2.5-Chat-7B | 64.3 | #1034 |
| 8 | Phi-3.5-mini-instruct | 63.7 | #1167 |
| 9 | granite-3.1-8B-instruct | 60.8 | #1069 |
| 10 | Yi-1.5-6B-Chat | 60.5 | #1259 |
| 11 | Falcon3-10B-Instruct | 54.3 | #1157 |
| 12 | EXAONE-3.5-2.4B-Instruct | 53.7 | #1391 |
| 13 | Falcon3-7B-Instruct | 52.4 | #1305 |
| 14 | Llama 3.2 3B Instruct | 50.2 | #1352 |
| 15 | Qwen 2.5 1.5B Instruct | 49.7 | #1407 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mobile-mmlu · How It Works · Data refreshed daily, snapshot 2026-10-20.