MMLU-by-task - Prehistory — leaderboard
Metric: Accuracy (%). Source: huggingface.co. 1257 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 2 70B Base | 83.64 |
| 2 | Llama 2 70B Chat | 82.72 |
| 3 | falcon-180B | 81.17 |
| 4 | StableBeluga2 | 79.63 |
| 5 | LLaMA-65B | 74.69 |
| 6 | Mistral-7B-v0.1 | 73.46 |
| 7 | Llama 2 70B Chat (HF) | 70.99 |
| 8 | Llama 2 70B Chat GPTQ | 70.37 |
| 9 | vicuna-33B-v1.3 | 68.52 |
| 10 | LLaMA-30B | 66.98 |
| 11 | OpenHermes-13B | 64.2 |
| 12 | vicuna-13B-v1.5 | 63.27 |
| 13 | MythoMax-L2-13B | 62.65 |
| 14 | falcon-40B Instruct | 62.04 |
| 15 | internlm-20B Chat | 61.73 |
Interactive version: theaggregate.ai/benchmark?slug=mmlu-by-task-prehistory · How the rankings work · Data refreshed daily, snapshot 2026-07-22.