MMLU: College Medicine

MMLU subject subset: Medical-school coursework, distinct from the licensing-exam-style Professional Medicine subject.

unassessed

This page is a discovery lead. Nobody has yet assessed it against the catalogue contract, so it carries no disposition. Absence of evidence here is not evidence of staleness.
Categoryknowledge
Subcategoryhealth
Page statusactive
Metricaccuracy
Directionhigher_is_better
Unit%
Dataset size173
Dataset licenceMIT
PublisherUC Berkeley (original); Center for AI Safety (current host)

What it measures

Medical-school coursework, distinct from the licensing-exam-style Professional Medicine subject. 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.

Task format

Four-option multiple-choice questions, graded on the single correct labelled option; commonly evaluated 5-shot, consistent with the rest of MMLU.

Models reporting this benchmark

These figures come from the model cards, which carry one collection date per card and no per-score attribution. They are shown as reported, not as verified evidence.
ModelProviderScoreCard as of
Meta Llama 3 70BMeta81.52024-07
Meta Llama 3 70B InstructMeta81.52026-04
Meta Llama 3 70B InstructNous Research81.52026-04
Mixtral 8x22B Instruct v0.1Mistral AI76.32026-04
Mixtral 8x7B Instruct v0.1Mistral AI75.72026-04
Yi 1.5 34B01.AI75.12026-04
Yi 1.5 34B Chat01.AI75.12026-04
Yi 1.5 34B Chat 16K01.AI75.12024-07
Yi 1.5 34B 32K01.AI74.62024-07
Yi 9B01.AI74.62024-07
Yi 1.5 9B Chat01.AI71.12024-07
Yi 34B 200K01.AI71.12024-07
Yi 1.5 9B 32K01.AI70.52024-07
gemma 7B itGoogle DeepMind69.92024-07
Mixtral 8x7B v0.1Mistral AI69.92026-04
Nous Hermes 2 Yi 34BNous Research69.92024-07
Yi 1.5 9B01.AI69.92024-07
Yi 34B Chat01.AI69.92024-07
Nous Hermes 2 Mixtral 8x7B DPONous Research69.42024-07
Yi 1.5 9B Chat 16K01.AI69.42024-07
Phi 3 mini 4K instructMicrosoft68.82024-07
Phi 3 mini 128K instructMicrosoft68.22024-07
Hermes 2 Theta Llama 3 8BNous Research66.52024-07
Nous Hermes 2 SOLAR 10.7BNous Research64.22024-07
Meta Llama 3 8BMeta63.62024-07
Meta Llama 3 8BNous Research63.62024-07
Meta Llama 3 8B InstructMeta63.62024-07
Meta Llama 3 8B InstructNous Research63.62024-07
Mistral 7B v0.3Mistral AI63.02024-07
mistral 7B v0.3 bnb 4bitUnsloth63.02024-07
Hermes 2 Pro Llama 3 8BNous Research61.82024-07
Yi 1.5 6B01.AI60.72024-07
Yi 1.5 6B Chat01.AI60.72024-07
Yi 6B01.AI60.72024-07
Yi 6B Chat01.AI60.72024-07
Mistral 7B Instruct v0.2Mistral AI59.02024-07
phi 2Microsoft59.02024-07
falcon 40BTII52.02024-07
deepseek llm 7B baseDeepSeek49.12024-07
deepseek llm 7B chatDeepSeek49.12024-07
Qwen2 1.5B InstructAlibaba / Qwen Team49.12024-07
Qwen2 0.5B InstructAlibaba / Qwen Team48.02024-07
gemma 2BGoogle DeepMind42.82024-07
chatglm2 6BZhipu AI42.22024-07
gemma 2B itGoogle DeepMind35.82024-07
deepseek coder 6.7B baseDeepSeek34.72024-07
deepseek coder 6.7B instructDeepSeek33.52024-07
OLMo 1B hfAllen AI27.22024-07
deepseek coder 1.3B baseDeepSeek24.32024-07
deepseek coder 1.3B instructDeepSeek24.32024-07

Data

This page as JSON · Edit on GitHub