MMLU-by-task - College Physics: leaderboard
The "College Physics" subtask of MMLU-by-task. A single MMLU subject, reporting per-task multiple-choice accuracy from the MMLU-by-task leaderboard across a wide range of academic and professional domains.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 1257 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | internlm-20B | 40.2 | #1296 |
| 2 | Mistral-7B-v0.1 | 39.22 | #1326 |
| 3 | StableBeluga2 | 39.22 | #1078 |
| 4 | Xwin-LM-70B-V0.1 | 38.24 | #982 |
| 5 | Llama 2 70B Base | 37.25 | #1025 |
| 6 | mpt-30B Instruct | 35.29 | #1470 |
| 7 | Chinese-Alpaca-2-13B | 35.29 | #1414 |
| 8 | CodeLlama-7B-Python-hf | 35.29 | #1608 |
| 9 | Llama 2 70B Chat (HF) | 33.33 | #1264 |
| 10 | CodeLlama-13B-Instruct-hf | 32.35 | #1532 |
| 11 | Llama 2 13B Chat Base | 31.37 | #1422 |
| 12 | vicuna-13B-v1.5 | 31.37 | #1392 |
| 13 | opt-1.3B | 31.37 | #1636 |
| 14 | MPT-30B-Chat | 31.37 | #1451 |
| 15 | Llama 2 70B Chat | 30.39 | #1237 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mmlu-by-task-college-physics · How It Works · Data refreshed daily, snapshot 2026-10-20.