MMLU subject subset: USMLE-style clinical vignettes on diagnosis, mechanism and management across medical specialties.
unassessed
| Category | knowledge |
|---|---|
| Subcategory | health |
| Page status | active |
| Metric | accuracy |
| Direction | higher_is_better |
| Unit | % |
| Dataset size | 272 |
| Dataset licence | MIT |
| Publisher | UC Berkeley (original); Center for AI Safety (current host) |
USMLE-style clinical vignettes: a patient history and findings followed by a question on diagnosis, mechanism or next step in management. Questions are four-option multiple-choice, drawn from the MMLU test set's "health" subcategory within the benchmark's "other" top-level group, and are graded on the single correct labelled option.
Four-option multiple-choice questions, graded on the single correct labelled option; commonly evaluated 5-shot, consistent with the rest of MMLU.
| Model | Provider | Score | Card as of |
|---|---|---|---|
| Meta Llama 3 70B | Meta | 89.0 | 2024-07 |
| Meta Llama 3 70B Instruct | Meta | 89.0 | 2026-04 |
| Meta Llama 3 70B Instruct | Nous Research | 89.0 | 2026-04 |
| Mixtral 8x22B Instruct v0.1 | Mistral AI | 88.6 | 2026-04 |
| Yi 1.5 34B | 01.AI | 84.9 | 2026-04 |
| Nous Hermes 2 Yi 34B | Nous Research | 83.1 | 2024-07 |
| Yi 1.5 34B 32K | 01.AI | 82.7 | 2024-07 |
| Mixtral 8x7B v0.1 | Mistral AI | 81.2 | 2026-04 |
| Yi 1.5 34B Chat | 01.AI | 81.2 | 2026-04 |
| Yi 1.5 34B Chat 16K | 01.AI | 81.2 | 2024-07 |
| Yi 34B 200K | 01.AI | 80.9 | 2024-07 |
| Mixtral 8x7B Instruct v0.1 | Mistral AI | 79.4 | 2026-04 |
| Nous Hermes 2 Mixtral 8x7B DPO | Nous Research | 78.7 | 2024-07 |
| Yi 34B Chat | 01.AI | 78.3 | 2024-07 |
| Phi 3 mini 4K instruct | Microsoft | 76.1 | 2024-07 |
| Nous Hermes 2 SOLAR 10.7B | Nous Research | 75.7 | 2024-07 |
| Phi 3 mini 128K instruct | Microsoft | 72.4 | 2024-07 |
| Meta Llama 3 8B | Meta | 72.1 | 2024-07 |
| Meta Llama 3 8B | Nous Research | 72.1 | 2024-07 |
| Meta Llama 3 8B Instruct | Meta | 71.7 | 2024-07 |
| Meta Llama 3 8B Instruct | Nous Research | 71.7 | 2024-07 |
| Yi 9B | 01.AI | 71.0 | 2024-07 |
| Yi 1.5 9B | 01.AI | 70.2 | 2024-07 |
| Hermes 2 Pro Llama 3 8B | Nous Research | 69.5 | 2024-07 |
| Hermes 2 Theta Llama 3 8B | Nous Research | 69.5 | 2024-07 |
| Yi 1.5 9B 32K | 01.AI | 69.5 | 2024-07 |
| Mistral 7B v0.3 | Mistral AI | 68.8 | 2024-07 |
| mistral 7B v0.3 bnb 4bit | Unsloth | 68.8 | 2024-07 |
| Yi 1.5 9B Chat 16K | 01.AI | 68.0 | 2024-07 |
| Yi 1.5 9B Chat | 01.AI | 67.6 | 2024-07 |
| Yi 6B | 01.AI | 67.3 | 2024-07 |
| Yi 6B Chat | 01.AI | 67.3 | 2024-07 |
| Yi 1.5 6B | 01.AI | 65.8 | 2024-07 |
| gemma 7B it | Google DeepMind | 63.2 | 2024-07 |
| Mistral 7B Instruct v0.2 | Mistral AI | 61.8 | 2024-07 |
| falcon 40B | TII | 61.0 | 2024-07 |
| Yi 1.5 6B Chat | 01.AI | 53.7 | 2024-07 |
| Qwen2 1.5B Instruct | Alibaba / Qwen Team | 50.4 | 2024-07 |
| phi 2 | Microsoft | 48.2 | 2024-07 |
| deepseek llm 7B base | DeepSeek | 46.7 | 2024-07 |
| deepseek llm 7B chat | DeepSeek | 46.7 | 2024-07 |
| deepseek coder 6.7B base | DeepSeek | 44.9 | 2024-07 |
| Qwen2 0.5B Instruct | Alibaba / Qwen Team | 44.9 | 2024-07 |
| deepseek coder 1.3B base | DeepSeek | 41.5 | 2024-07 |
| deepseek coder 1.3B instruct | DeepSeek | 41.5 | 2024-07 |
| OLMo 1B hf | Allen AI | 40.4 | 2024-07 |
| chatglm2 6B | Zhipu AI | 35.3 | 2024-07 |
| gemma 2B | Google DeepMind | 35.3 | 2024-07 |
| deepseek coder 6.7B instruct | DeepSeek | 34.9 | 2024-07 |
| gemma 2B it | Google DeepMind | 20.2 | 2024-07 |