registry.csv

benchmark,evaluation_suites,frontier_model_reports_by,in_suite_meant_to_resemble_practice,used_in_frontier_model_reports,items_sampled,items_labelled,items_usable,split,version,license,source,rendering_rule,share_of_rcq_task_mix
ACI-Bench,MEDIC; MedHELM; Medmarks,,True,False,120,120,120,"test1 + test2 + test3 (MEDIQA-CHAT 2023 Task B, Task C; MEDIQA-SUM 2023 Task C), 40 encounters each",b909b2bb9cf1d19de08df15cddde7bd0179665e4,CC BY 4.0 (README),https://raw.githubusercontent.com/wyim/aci-bench/b909b2bb9cf1d19de08df15cddde7bd0179665e4/README.md; https://raw.githubusercontent.com/wyim/aci-bench/b909b2bb9cf1d19de08df15cddde7bd0179665e4/data/challenge_data/clinicalnlp_taskB_test1.csv; https://raw.githubusercontent.com/wyim/aci-bench/b909b2bb9cf1d19de08df15cddde7bd0179665e4/data/challenge_data/clinicalnlp_taskB_test1_metadata.csv; and 4 more files listed in registry/manifests/aci_bench.json,"Full MedHELM prompt exactly as HELM builds it (generation adapter, zero-shot): instructions, blank line, ""Conversation: Doctor-patient dialogue:"" + dialogue, newline, final cue ""Clinical Note:"" (dialogues verified identical to HELM-pinned challenge_data_json at e75b383).",0.351
ADE corpus,BRIDGE; MedS-Bench,,True,False,23,23,23,MedS-Bench test file(s); MedS-Bench-SPLIT has 23 (identical),HF Henrychur/MedS-Bench 6945b14e0f408082aecb6a77a185b9ebdd5fb413; HF Henrychur/MedS-Bench-SPLIT bae9b49bb23ffcd5b1b64345f021d4d5b803b503 (GitHub MAGIC-AI4Med/MedS-Ins 185cfc04a27023983e48b357a6f2fd525049b9af),"MedS-Ins/MedS-Bench GitHub LICENSE: ""CC BY-SA"" (MAGIC consortium); HF dataset cards declare no license; task-file Instance License: Unknown; source ADE-Corpus-V2 (Gurulingappa et al. 2012)",https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/Information_extraction/task29_drug_dose_extraction.json; https://huggingface.co/datasets/Henrychur/MedS-Bench-SPLIT/resolve/bae9b49bb23ffcd5b1b64345f021d4d5b803b503/Information_extraction/task29_drug_dose_extraction.json,"instruction-style (MedS-Bench): the task Definition, a blank line, then the instance input (few-shot examples and outputs omitted)",0.348
BC4CHEMD,MedS-Bench,,False,False,200,200,26203,"MedS-Bench test file(s); MedS-Bench-SPLIT has 1,500 (random subset)",HF Henrychur/MedS-Bench 6945b14e0f408082aecb6a77a185b9ebdd5fb413; HF Henrychur/MedS-Bench-SPLIT bae9b49bb23ffcd5b1b64345f021d4d5b803b503 (GitHub MAGIC-AI4Med/MedS-Ins 185cfc04a27023983e48b357a6f2fd525049b9af),"MedS-Ins/MedS-Bench GitHub LICENSE: ""CC BY-SA"" (MAGIC consortium); HF dataset cards declare no license; task-file Instance License: none given; source BC4CHEMD/CHEMDNER test via cambridgeltl/MTL-Bioinformatics-2016",https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/NER/task125_test_bc4chem_named_enetity_recognition.json; https://huggingface.co/datasets/Henrychur/MedS-Bench-SPLIT/resolve/bae9b49bb23ffcd5b1b64345f021d4d5b803b503/NER/task125_test_bc4chem_named_enetity_recognition.json,"instruction-style (MedS-Bench): the task Definition, a blank line, then the instance input (few-shot examples and outputs omitted)",0.206
BC5CDR,ClinicBench; MedS-Bench,,True,False,200,200,9574,"MedS-Bench test file(s); MedS-Bench-SPLIT has 1,500 + 1,500 (random subsets)",HF Henrychur/MedS-Bench 6945b14e0f408082aecb6a77a185b9ebdd5fb413; HF Henrychur/MedS-Bench-SPLIT bae9b49bb23ffcd5b1b64345f021d4d5b803b503 (GitHub MAGIC-AI4Med/MedS-Ins 185cfc04a27023983e48b357a6f2fd525049b9af),"MedS-Ins/MedS-Bench GitHub LICENSE: ""CC BY-SA"" (MAGIC consortium); HF dataset cards declare no license; task-file Instance License: none given; source BC5CDR test via cambridgeltl/MTL-Bioinformatics-2016",https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/NER/task126_test_bc5chem_named_enetity_recognition.json; https://huggingface.co/datasets/Henrychur/MedS-Bench-SPLIT/resolve/bae9b49bb23ffcd5b1b64345f021d4d5b803b503/NER/task126_test_bc5chem_named_enetity_recognition.json; https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/NER/task127_test_bc5disease_named_enetity_recognition.json; and 1 more files listed in registry/manifests/bc5cdr.json,"instruction-style (MedS-Bench): the task Definition, a blank line, then the instance input (few-shot examples and outputs omitted)",0.253
BioASQ,MIRAGE,Anthropic,False,True,200,200,617,MIRAGE BioASQ-Y/N: yes/no questions from BioASQ Task B gold test sets 2019-2023 (Tasks 7B-11B),392943af99cd94cafd50a0de2e7fca24bbf65494,MIRAGE repo: NCBI public-domain notice; underlying BioASQ Task B data subject to BioASQ terms,https://raw.githubusercontent.com/Teddy-XiongGZ/MIRAGE/392943af99cd94cafd50a0de2e7fca24bbf65494/README.md; https://raw.githubusercontent.com/Teddy-XiongGZ/MIRAGE/392943af99cd94cafd50a0de2e7fca24bbf65494/LICENSE; https://raw.githubusercontent.com/Teddy-XiongGZ/MIRAGE/392943af99cd94cafd50a0de2e7fca24bbf65494/benchmark.json,"Multiple choice: question, then options one per line with their source letters (A. yes / B. no).",0.374
BioLORD concept explanation,MedS-Bench,,False,False,200,200,1000,"test (MedS-Bench test file, all instances)",6945b14e0f408082aecb6a77a185b9ebdd5fb413,MedS-Bench card states no license; BioLORD-Dataset is tagged license 'other' on the Hub (UMLS/SNOMED-derived).,https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/Explanation/task50_biolord_explanation.json; https://huggingface.co/datasets/Henrychur/MedS-Bench-SPLIT/resolve/bae9b49bb23ffcd5b1b64345f021d4d5b803b503/Explanation/task50_biolord_explanation.json; https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/README.md,"instruction-style: the MedS-Bench task Definition (instruction), a blank line, then the instance input.",0.374
BrainMRI-AIS,BRIDGE,,True,False,200,200,263,test,41159677fba2454e032574625e24b24d29a17684,see BRIDGE-Open dataset card and the original dataset licence,https://huggingface.co/datasets/YLab-Open/BRIDGE-Open/resolve/41159677fba2454e032574625e24b24d29a17684/Dataset/BrainMRI-AIS.SFT.json,"instruction-style: the BRIDGE instruction, a blank line, then the input.",0.032
CareQA,Medmarks,,False,False,200,200,5621,test (only split; closed-ended English file CareQA_en.json),1d976cc544ccb9dfc96e14439e6840e8a810cec5,apache-2.0,https://huggingface.co/datasets/HPAI-BSC/CareQA/resolve/1d976cc544ccb9dfc96e14439e6840e8a810cec5/CareQA_en.json; https://huggingface.co/datasets/HPAI-BSC/CareQA/resolve/1d976cc544ccb9dfc96e14439e6840e8a810cec5/CareQA_en_open.json; https://huggingface.co/datasets/HPAI-BSC/CareQA/resolve/1d976cc544ccb9dfc96e14439e6840e8a810cec5/README.md,"multiple choice: question, then options op1-op4 one per line lettered A-D (source has no letters).",0.47
ClinicalNotes-UPMC,BRIDGE,,True,False,200,200,229,test,41159677fba2454e032574625e24b24d29a17684,see BRIDGE-Open dataset card and the original dataset licence,https://huggingface.co/datasets/YLab-Open/BRIDGE-Open/resolve/41159677fba2454e032574625e24b24d29a17684/Dataset/ClinicalNotes-UPMC.SFT.json,"instruction-style: the BRIDGE instruction, a blank line, then the input.",0.226
DDXPlus,MedS-Bench,,False,False,200,200,134428,"test (MedS-Bench test file, all instances)",6945b14e0f408082aecb6a77a185b9ebdd5fb413,MedS-Bench card states no license; Original DDXPlus license not checked here.,https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/Diagnosis/task130_DDXPlus_text_classification_test.json; https://huggingface.co/datasets/Henrychur/MedS-Bench-SPLIT/resolve/bae9b49bb23ffcd5b1b64345f021d4d5b803b503/Diagnosis/task130_DDXPlus_text_classification_test.json; https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/README.md,"instruction-style: the MedS-Bench task Definition (instruction), a blank line, then the instance input.",0.037
Drug Interaction for Emerging Drugs,ClinicBench,,True,False,200,200,200,whole set (no split; evaluation-only dataset),57138be8a73f09bef8980f849789baf57570c5cf,Apache-2.0 (ClinicBench repository LICENSE),https://raw.githubusercontent.com/AI-in-Health/ClinicBench/57138be8a73f09bef8980f849789baf57570c5cf/Drug-Interaction-for-Emerging-Drugs.csv; https://raw.githubusercontent.com/AI-in-Health/ClinicBench/57138be8a73f09bef8980f849789baf57570c5cf/README.md; https://raw.githubusercontent.com/AI-in-Health/ClinicBench/57138be8a73f09bef8980f849789baf57570c5cf/LICENSE,"Item text as released (the yes/no question); the ClinicBench Table 8 prompt was not applied because it contains an unpublished placeholder ""(introduce what each Moderna COVID-19 Vaccine is ...)"".",0.141
EBM-NLP PICO extraction,MedS-Bench,,False,False,129,129,129,MedS-Bench test file(s); MedS-Bench-SPLIT has 43 + 43 + 43 (identical),HF Henrychur/MedS-Bench 6945b14e0f408082aecb6a77a185b9ebdd5fb413; HF Henrychur/MedS-Bench-SPLIT bae9b49bb23ffcd5b1b64345f021d4d5b803b503 (GitHub MAGIC-AI4Med/MedS-Ins 185cfc04a27023983e48b357a6f2fd525049b9af),"MedS-Ins/MedS-Bench GitHub LICENSE: ""CC BY-SA"" (MAGIC consortium); HF dataset cards declare no license; task-file Instance License: Unknown; source ""pico"" (github.com/Markus-Zlabinger/pico-annotation, crowd re-annotation of EBM-NLP sentences)",https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/Information_extraction/task1_participant_extraction.json; https://huggingface.co/datasets/Henrychur/MedS-Bench-SPLIT/resolve/bae9b49bb23ffcd5b1b64345f021d4d5b803b503/Information_extraction/task1_participant_extraction.json; https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/Information_extraction/task2_intervention_extraction.json; and 3 more files listed in registry/manifests/ebm_nlp_pico_extraction.json,"instruction-style (MedS-Bench): the task Definition, a blank line, then the instance input (few-shot examples and outputs omitted)",0.206
EBMSummariser justification verification,MedS-Bench,,False,False,110,110,110,"test (MedS-Bench test file, all instances)",6945b14e0f408082aecb6a77a185b9ebdd5fb413,MedS-Bench card states no license; EBMSummariser corpus license not checked.,https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/Fact_verication/task100_ebms_answer_vertification.json; https://huggingface.co/datasets/Henrychur/MedS-Bench-SPLIT/resolve/bae9b49bb23ffcd5b1b64345f021d4d5b803b503/Fact_verication/task100_ebms_answer_vertification.json; https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/README.md,"instruction-style: the MedS-Bench task Definition (instruction), a blank line, then the instance input.",0.566
EHRSQL,MEDIC; MedHELM,,True,False,200,200,3499,"test (mimic_iii/test.json + eicu/test.json), dataset v1.5.1",01241a3c55bda895561c76923251ea2421d2e892,CC BY 4.0 (LICENSE); databases derived from MIMIC-III and eICU (PhysioNet credentialed),https://raw.githubusercontent.com/glee4810/EHRSQL/01241a3c55bda895561c76923251ea2421d2e892/README.md; https://raw.githubusercontent.com/glee4810/EHRSQL/01241a3c55bda895561c76923251ea2421d2e892/LICENSE; https://raw.githubusercontent.com/glee4810/EHRSQL/01241a3c55bda895561c76923251ea2421d2e892/dataset/ehrsql/CHANGELOG.md; and 2 more files listed in registry/manifests/ehrsql.json,"Item text as released: the natural-language question only (no schema, no SQL).",0.774
First Do NOHARM,MAST,,False,False,200,200,330,open subset: 30 base cases x 11 variants (1 base + 10 perturbations),57a12c5490f3a7a6b0a6ce4e0d49f8e393ff49ee,Data CC BY 4.0 (donoharm/LICENSE-DATA); code MIT,https://codeload.github.com/ARISENetwork/mast/tar.gz/57a12c5490f3a7a6b0a6ce4e0d49f8e393ff49ee,"Item text as released: the case prompt (MAST NOHARM protocol is unprompted, empty template).",0.24
Hallmarks of Cancer,ClinicBench; MedS-Bench,,True,False,158,158,158,MedS-Bench test file(s); MedS-Bench-SPLIT has 158 (identical),HF Henrychur/MedS-Bench 6945b14e0f408082aecb6a77a185b9ebdd5fb413; HF Henrychur/MedS-Bench-SPLIT bae9b49bb23ffcd5b1b64345f021d4d5b803b503 (GitHub MAGIC-AI4Med/MedS-Ins 185cfc04a27023983e48b357a6f2fd525049b9af),"MedS-Ins/MedS-Bench GitHub LICENSE: ""CC BY-SA"" (MAGIC consortium); HF dataset cards declare no license; task-file Instance License: none given; source github.com/sb895/Hallmarks-of-Cancer",https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/Text_classification/task106_hoc_text_classification.json; https://huggingface.co/datasets/Henrychur/MedS-Bench-SPLIT/resolve/bae9b49bb23ffcd5b1b64345f021d4d5b803b503/Text_classification/task106_hoc_text_classification.json,"instruction-style (MedS-Bench): the task Definition, a blank line, then the instance input (few-shot examples and outputs omitted)",0.206
HEAD-QA,MMedBench; MedHELM; MedS-Bench; Medmarks,,True,False,200,200,2674,"test (HEAD_EN/test_HEAD_EN.json, English version)",d444ead98dd50d7949703c1662c7880a5c1ea77d,mit,https://huggingface.co/datasets/dvilares/head_qa/resolve/d444ead98dd50d7949703c1662c7880a5c1ea77d/README.md; https://huggingface.co/datasets/dvilares/head_qa/resolve/d444ead98dd50d7949703c1662c7880a5c1ea77d/data/head-qa-es-en-pdfs.zip#HEAD_EN/test_HEAD_EN.json,"multiple choice: qtext, then answers one per line in aid order lettered A-E (source numbers them 1-5); questions with a non-empty 'image' field are excluded as image items in addition to the regex.",0.445
HealthBench,HealthBench; MEDIC; Medmarks,Alibaba; Anthropic; OpenAI,True,True,200,200,4412,full eval set (oss_eval),"Azure blob ETag 0x8DD9176E8C2022D, Last-Modified 2025-05-12, Content-MD5 WG5ihzNmgj88RcJMS+YHuA== (verified); referenced by openai/simple-evals 652c89d0ca9df547706735883097e9537d40dc47 healthbench_eval.py",MIT (openai/simple-evals; HF openai/healthbench card: mit),https://openaipublic.blob.core.windows.net/simple-evals/healthbench/2025-05-07-06-14-12_oss_eval.jsonl,"conversation: the first user turn only; language identified with lingua-language-detector 2.1.1 (LanguageDetectorBuilder.from_all_languages().with_preloaded_language_models().build()); an item is non-English only if compute_language_confidence_values on the first user turn gives a top language other than ENGLISH with confidence >= 0.80, every other item (including short ones) counts as English",0.621
HealthBench Professional,HealthBench,Anthropic,True,True,200,200,461,public eval set (OpenAI keeps a private held-out set),blob Last-Modified 2026-04-22; assets.zip sha256 a5286ead1f03d178c7f7b134a5ac16944998428303012f0921814701e0f3995e,MIT (c) 2026 OpenAI; README asks that examples not be revealed in plain text or images online,https://openaipublic.blob.core.windows.net/simple-evals/healthbench_professional/assets.zip,"conversation: the first user turn only; language identified with lingua-language-detector 2.1.1 (LanguageDetectorBuilder.from_all_languages().with_preloaded_language_models().build()); an item is non-English only if compute_language_confidence_values on the first user turn gives a top language other than ENGLISH with confidence >= 0.80, every other item (including short ones) counts as English",0.766
HealthCareMagic-100k,BRIDGE,,True,False,200,200,11195,test,41159677fba2454e032574625e24b24d29a17684,see BRIDGE-Open dataset card and the original dataset licence,https://huggingface.co/datasets/YLab-Open/BRIDGE-Open/resolve/41159677fba2454e032574625e24b24d29a17684/Dataset/HealthCareMagic-100k.SFT.json,"instruction-style: the BRIDGE instruction, a blank line, then the input.",0.356
HealthSearchQA,MEDIC; MultiMedQA,Google,False,True,200,200,3156,"sheet ""All HealthSearchQA Questions"" (no split)",Nature ESM file MOESM6 (article published 2023-07-12),CC BY 4.0 (article licence),https://media.springernature.com/original/springer-static/esm/art%3A10.1038%2Fs41586-023-06291-2/MediaObjects/41586_2023_6291_MOESM6_ESM.xlsx,"otherwise: the consumer question as released (sheet ""All HealthSearchQA Questions"")",0.289
Human Disease Ontology entity explanation,MedS-Bench,,False,False,200,200,1115,"test (MedS-Bench test file, all instances)",6945b14e0f408082aecb6a77a185b9ebdd5fb413,MedS-Bench card states no license; Underlying Human Disease Ontology license not checked here.,https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/Explanation/task46_do_entity_explanation.json; https://huggingface.co/datasets/Henrychur/MedS-Bench-SPLIT/resolve/bae9b49bb23ffcd5b1b64345f021d4d5b803b503/Explanation/task46_do_entity_explanation.json; https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/README.md,"instruction-style: the MedS-Bench task Definition (instruction), a blank line, then the instance input.",0.153
icliniq-10k,BRIDGE,,True,False,200,200,713,test,41159677fba2454e032574625e24b24d29a17684,see BRIDGE-Open dataset card and the original dataset licence,https://huggingface.co/datasets/YLab-Open/BRIDGE-Open/resolve/41159677fba2454e032574625e24b24d29a17684/Dataset/icliniq-10k.SFT.json,"instruction-style: the BRIDGE instruction, a blank line, then the input.",0.356
IU X-ray report summarization,ClinicBench,,True,False,200,200,2627,whole report set (no official split); ClinicBench uses a 10% test split of 341 reports that is not released,"Open-i archive, Last-Modified 2017-06-06 (no version id)",CC BY-NC-ND 4.0 (licenseURL in every report XML),https://openi.nlm.nih.gov/imgs/collections/NLMCXR_reports.tgz,"template: ClinicBench radiology report summarization prompt (paper Table 8, adapted from Tu et al. 2023) filled with the report FINDINGS (whitespace collapsed); reports without FINDINGS and IMPRESSION are not items",0.351
LiveQA,MultiMedQA,Google,False,True,104,104,104,TREC 2017 LiveQA Medical test set,725a2d914e9794afbb897ac2006e3b0c6a7834c4,CC BY 4.0 (Readme.txt),https://raw.githubusercontent.com/abachaa/LiveQA_MedicalTask_TREC2017/725a2d914e9794afbb897ac2006e3b0c6a7834c4/Readme.txt; https://raw.githubusercontent.com/abachaa/LiveQA_MedicalTask_TREC2017/725a2d914e9794afbb897ac2006e3b0c6a7834c4/TestDataset/TREC-2017-LiveQA-Medical-Test-Questions-w-summaries.xml; https://raw.githubusercontent.com/abachaa/LiveQA_MedicalTask_TREC2017/725a2d914e9794afbb897ac2006e3b0c6a7834c4/TestDataset/TREC-2017-LiveQA-Medical-Test.xml,"Item text as released: the original consumer question, SUBJECT line then MESSAGE (SUBJECT omitted when empty).",0.419
LongHealth,CLUE; Medmarks,,False,False,200,200,400,whole benchmark (benchmark_v5.json; no split),479922279ce8fa528fe2cd87d854c6a2f8365d21,Apache-2.0 (LICENSE),https://raw.githubusercontent.com/kbressem/LongHealth/479922279ce8fa528fe2cd87d854c6a2f8365d21/README.md; https://raw.githubusercontent.com/kbressem/LongHealth/479922279ce8fa528fe2cd87d854c6a2f8365d21/data/README.md; https://raw.githubusercontent.com/kbressem/LongHealth/479922279ce8fa528fe2cd87d854c6a2f8365d21/LICENSE; https://raw.githubusercontent.com/kbressem/LongHealth/479922279ce8fa528fe2cd87d854c6a2f8365d21/data/benchmark_v5.json,"Vignette + question: all of the patient's documents in text_0..text_n order joined by the repo separator (""--------------""), a blank line, then the question and options A-E (no shuffling or truncation, unlike the repo runner).",0.282
M-ARC,Medmarks,,False,False,99,99,99,test,e69ae5f92b3a5d550cd98adcd5e5bef281daf1c2,apache-2.0,https://huggingface.co/datasets/mkieffer/M-ARC/resolve/e69ae5f92b3a5d550cd98adcd5e5bef281daf1c2/data/test-00000-of-00001.parquet; https://huggingface.co/datasets/mkieffer/M-ARC/resolve/e69ae5f92b3a5d550cd98adcd5e5bef281daf1c2/README.md,"multiple choice: question, then the source's lettered options (A-G) one per line.",0.232
Med-HALT,Medmarks,,False,False,200,160,58747,none (whole set: 3 reasoning + 4 memory test configs),d1e8d04295742df7abc485062344fe328e141395,apache-2.0,https://huggingface.co/datasets/openlifescienceai/Med-HALT/resolve/d1e8d04295742df7abc485062344fe328e141395/IR_abstract2pubmedlink/IR_abstract2pubmedlink.csv; https://huggingface.co/datasets/openlifescienceai/Med-HALT/resolve/d1e8d04295742df7abc485062344fe328e141395/IR_pmid2title/IR_pmid2title.csv; https://huggingface.co/datasets/openlifescienceai/Med-HALT/resolve/d1e8d04295742df7abc485062344fe328e141395/IR_pubmedlink2title/IR_pubmedlink2title.csv; and 5 more files listed in registry/manifests/med_halt.json,"template-defined: Med-HALT's published prompt for the item's test (version v0, the version used by the repository's inference), zero-shot (the repository's 2 few-shot examples are drawn with an unseeded random.sample and are omitted), filled exactly as the repository's prompt utilities do: instruction + '\n' + output format + '\n' + 'Input: ' + Python dict of the item fields + '\nOutput: '.",0.479
Med-PaLM 2 adversarial questions,,Google,False,True,182,182,182,"OMAQ (= Med-PaLM 2 ""Adversarial (Health equity)"")",Figshare 10.6084/m9.figshare.26133973.v1 (published 2024-09-24); OMAQ file md5 a65ae305358e9f88f1360b5b5e388865 (verified),CC BY 4.0 (Figshare item and LICENSE),https://ndownloader.figshare.com/files/48320776; https://ndownloader.figshare.com/files/48320803; https://ndownloader.figshare.com/files/48320782,otherwise: the question text as released (EquityMedQA OMAQ question_text),0.58
Medbullets,MedHELM; Medmarks,,True,False,185,185,185,whole set (308 questions; op4 and op5 versions of the same questions),013b9447f2a2426155c0de02a42e03d4540a3501,none stated (no LICENSE file in repo),https://raw.githubusercontent.com/HanjieChen/ChallengeClinicalQA/013b9447f2a2426155c0de02a42e03d4540a3501/README.md; https://raw.githubusercontent.com/HanjieChen/ChallengeClinicalQA/013b9447f2a2426155c0de02a42e03d4540a3501/medbullets/medbullets_op4.csv; https://raw.githubusercontent.com/HanjieChen/ChallengeClinicalQA/013b9447f2a2426155c0de02a42e03d4540a3501/medbullets/medbullets_op5.csv,"Multiple choice (op5 file, as MedHELM uses): question, then options A-E one per line.",0.316
MedCalc-Bench,MEDIC; MedHELM; Medmarks,,True,False,200,200,1099,test (test_data_11_18_final.csv; train_data not downloaded),83a1a56d6b53f8d4c0e03f0ba9699becf524a595,cc-by-sa-4.0,https://huggingface.co/datasets/ncbi/MedCalc-Bench/resolve/83a1a56d6b53f8d4c0e03f0ba9699becf524a595/test_data_11_18_final.csv; https://huggingface.co/datasets/ncbi/MedCalc-Bench/resolve/83a1a56d6b53f8d4c0e03f0ba9699becf524a595/README.md,"vignette + question: Patient Note, a blank line, then Question.",0.29
MedCaseReasoning,Medmarks,,False,False,200,200,897,"test (data/test parquet; train, val and the 205 MB core files not downloaded)",469a5365bc534b5b2b9cfbc52b2ef2a10f43a339,mit,https://huggingface.co/datasets/zou-lab/MedCaseReasoning/resolve/469a5365bc534b5b2b9cfbc52b2ef2a10f43a339/data/test-00000-of-00001.parquet; https://huggingface.co/datasets/zou-lab/MedCaseReasoning/resolve/469a5365bc534b5b2b9cfbc52b2ef2a10f43a339/README.md,template-defined: MedCaseReasoning's published Diagnostic Question Template (zero-shot) with case_prompt as {case_presentation}.,0.037
MedConceptsQA,Medmarks,,False,False,200,200,819772,"test (config 'all', 2 parquet shards; dev few-shot files not downloaded)",98c30d83762e51a397c9a7b0eeee6722e751da17,apache-2.0,https://huggingface.co/datasets/ofir408/MedConceptsQA/resolve/98c30d83762e51a397c9a7b0eeee6722e751da17/all/test-00000-of-00002.parquet; https://huggingface.co/datasets/ofir408/MedConceptsQA/resolve/98c30d83762e51a397c9a7b0eeee6722e751da17/all/test-00001-of-00002.parquet; https://huggingface.co/datasets/ofir408/MedConceptsQA/resolve/98c30d83762e51a397c9a7b0eeee6722e751da17/README.md,multiple choice as released: the question field already holds the stem and the four options one per line lettered A-D.,0.01
MedDialog,MedHELM; Medmarks,,True,False,200,200,24711,"test (healthcaremagic 22,642 + icliniq 3,108)","CodaLab bundle 0x82f0c47f6d3e4462ae9ef8ea39eebe64 (immutable, created 2022-08-23); HELM 63754d05db6f874e41a395880fb573890a13e791",none stated on the bundle; MedDialog notes all copyrights of the dialogues belong to healthcaremagic.com and icliniq.com,https://worksheets.codalab.org/rest/bundles/0x82f0c47f6d3e4462ae9ef8ea39eebe64/contents/blob/healthcaremagic/test.json; https://worksheets.codalab.org/rest/bundles/0x82f0c47f6d3e4462ae9ef8ea39eebe64/contents/blob/icliniq/test.json; https://raw.githubusercontent.com/stanford-crfm/helm/63754d05db6f874e41a395880fb573890a13e791/src/helm/benchmark/scenarios/med_dialog_scenario.py,"template: MedHELM med_dialog prompt (HELM generation adapter: instruction, blank line, ""Patient-Doctor: <dialogue>"", then the output cue ""Summary:""); healthcaremagic and icliniq pooled",0.351
MEDEC,MEDIC; MedHELM; Medmarks,,True,False,200,200,597,MEDEC-MS test (597); MEDEC-UW test (328) is request-only,e9173f861b39706946b6a127394992c00535b1e3,CC BY 4.0 (README),https://raw.githubusercontent.com/abachaa/MEDEC/e9173f861b39706946b6a127394992c00535b1e3/README.md; https://raw.githubusercontent.com/abachaa/MEDEC/e9173f861b39706946b6a127394992c00535b1e3/MEDEC-MS/MEDEC-MS-TestSet-with-GroundTruth-and-ErrorType.csv,"Full MedHELM prompt exactly as HELM builds it (generation adapter, zero-shot): instructions, blank line, ""Clinical Note: "" + sentence-numbered text, newline, final cue ""Answer:"" (test CSV identical to HELM-pinned commit 49c59dc).",0.288
MedExQA,Medmarks,,False,False,200,200,940,test (5 specialty TSVs; dev not downloaded),3a9f80d6de2c354956d7e86d323214b92d547d6e,cc-by-nc-sa-4.0,https://huggingface.co/datasets/bluesky333/MedExQA/resolve/3a9f80d6de2c354956d7e86d323214b92d547d6e/test/biomedical_engineer_test.tsv; https://huggingface.co/datasets/bluesky333/MedExQA/resolve/3a9f80d6de2c354956d7e86d323214b92d547d6e/test/clinical_laboratory_scientist_test.tsv; https://huggingface.co/datasets/bluesky333/MedExQA/resolve/3a9f80d6de2c354956d7e86d323214b92d547d6e/test/clinical_psychologist_test.tsv; and 3 more files listed in registry/manifests/medexqa.json,"multiple choice: question (column 0), then options (columns 1-4) one per line lettered A-D.",0.288
MedHallu,MedHELM; Medmarks,,True,False,200,200,10000,pqa_labeled + pqa_artificial (both named 'train' on the Hub; the dataset is evaluation-only),515060458a945c633debc6fd5baac7764416b724,MIT (stated in the dataset card; no license tag on the Hub),https://huggingface.co/datasets/UTAustin-AIHealth/MedHallu/resolve/515060458a945c633debc6fd5baac7764416b724/pqa_labeled/train-00000-of-00001.parquet; https://huggingface.co/datasets/UTAustin-AIHealth/MedHallu/resolve/515060458a945c633debc6fd5baac7764416b724/pqa_artificial/train-00000-of-00001.parquet; https://huggingface.co/datasets/UTAustin-AIHealth/MedHallu/resolve/515060458a945c633debc6fd5baac7764416b724/README.md,"template-defined: MedHallu's published detection prompt with world knowledge, filled exactly as in the script (system prompt, a space, then 'World Knowledge: <Knowledge list>', 'Question:', 'Answer: <candidate>' and the judgement cue). The candidate answer is the Ground Truth or the Hallucinated Answer, chosen per item by random.Random(20261004).randint(0, 1) in file order (the script uses an unseeded random.randint(0, 1)).",0.206
MedicationQA,MEDIC; MedHELM; Medmarks; MultiMedQA,Google,True,True,200,200,651,whole set (no split),20363d1913d535a5cac017812fdbf335259a0105,CC BY 4.0 (README),https://raw.githubusercontent.com/abachaa/Medication_QA_MedInfo2019/20363d1913d535a5cac017812fdbf335259a0105/README.md; https://raw.githubusercontent.com/abachaa/Medication_QA_MedInfo2019/20363d1913d535a5cac017812fdbf335259a0105/MedInfo2019-QA-Medications.xlsx,Item text as released: the consumer question.,0.301
MEDIQA 2019 Task 2 RQE,BRIDGE,,True,False,200,200,230,test,41159677fba2454e032574625e24b24d29a17684,see BRIDGE-Open dataset card and the original dataset licence,https://huggingface.co/datasets/YLab-Open/BRIDGE-Open/resolve/41159677fba2454e032574625e24b24d29a17684/Dataset/MEDIQA 2019-RQE.SFT.json,"instruction-style: the BRIDGE instruction, a blank line, then the input.",0.258
MEDIQA-QA,MedHELM,,True,False,148,148,148,MEDIQA 2019 Task 3 (QA) test set with labels,32311a139b583a9ccec133b3f8a21873d4ff3561,CC BY 4.0 (README),https://raw.githubusercontent.com/abachaa/MEDIQA2019/32311a139b583a9ccec133b3f8a21873d4ff3561/README.md; https://raw.githubusercontent.com/abachaa/MEDIQA2019/32311a139b583a9ccec133b3f8a21873d4ff3561/MEDIQA_Task3_QA/Task3_README.txt; https://raw.githubusercontent.com/abachaa/MEDIQA2019/32311a139b583a9ccec133b3f8a21873d4ff3561/MEDIQA_Task3_QA/MEDIQA2019-Task3-QA-TestSet-wLabels.xml,Item text as released: the consumer question alone (candidate answers omitted).,0.353
MedMCQA,ClinicBench; MEDIC; MIRAGE; MedHELM; MedS-Bench; Medmarks; MultiMedQA,Anthropic; Google; OpenAI,True,True,200,200,4178,validation (dev) - the split with answer keys that is used for evaluation; test also downloaded,91c6572c454088bf71b679ad90aa8dffcd0d5868,apache-2.0,https://huggingface.co/datasets/openlifescienceai/medmcqa/resolve/91c6572c454088bf71b679ad90aa8dffcd0d5868/data/validation-00000-of-00001.parquet; https://huggingface.co/datasets/openlifescienceai/medmcqa/resolve/91c6572c454088bf71b679ad90aa8dffcd0d5868/data/test-00000-of-00001.parquet; https://huggingface.co/datasets/openlifescienceai/medmcqa/resolve/91c6572c454088bf71b679ad90aa8dffcd0d5868/README.md,"multiple choice: question, then opa-opd one per line lettered A-D.",0.475
MedQA,ClinicBench; MEDIC; MIRAGE; MMedBench; MedHELM; MedS-Bench; Medmarks; MultiMedQA,Anthropic; Google; OpenAI,True,True,200,200,1273,US (English) 4-option test,GitHub jind11/MedQA 27b02f66aac217933c9648a06f82e9f720377925; HF GBaker 0fb93dd23a7339b6dcd27e241cb9b5eca62d4d18; HF bigbio 484a6c066fe8e75c83edea0c88b5169316714fcd; MIRAGE 392943af99cd94cafd50a0de2e7fca24bbf65494,MIT (jind11/MedQA LICENSE); GBaker card CC BY 4.0; bigbio card unknown,https://huggingface.co/datasets/GBaker/MedQA-USMLE-4-options/resolve/0fb93dd23a7339b6dcd27e241cb9b5eca62d4d18/phrases_no_exclude_test.jsonl; https://huggingface.co/datasets/bigbio/med_qa/resolve/484a6c066fe8e75c83edea0c88b5169316714fcd/med_qa_en_4options_source/test-00000-of-00001.parquet; https://raw.githubusercontent.com/jind11/MedQA/27b02f66aac217933c9648a06f82e9f720377925/README.md; and 2 more files listed in registry/manifests/medqa.json,"Multiple choice: question, then options A-D one per line.",0.387
MedR-Bench,Medmarks,,False,False,200,200,1453,whole benchmark: diagnosis (957 cases) + treatment (496 cases),ff60ab440afd2f2bc0c603b3a65d715ec83138a7,"CC BY-SA (LICENSE, ""Copyright (c) 2024 MAGIC"")",https://raw.githubusercontent.com/MAGIC-AI4Med/MedRBench/ff60ab440afd2f2bc0c603b3a65d715ec83138a7/README.md; https://raw.githubusercontent.com/MAGIC-AI4Med/MedRBench/ff60ab440afd2f2bc0c603b3a65d715ec83138a7/LICENSE; https://raw.githubusercontent.com/MAGIC-AI4Med/MedRBench/ff60ab440afd2f2bc0c603b3a65d715ec83138a7/data/MedRBench/diagnosis_957_cases_with_rare_disease_491.json; and 4 more files listed in registry/manifests/medr_bench.json,Repository prompt template filled exactly: oracle_diagnose.txt (diagnosis cases) or treatment_plan_prompt.txt (treatment cases) with {case} = generate_case.case_summary.,0.099
MedSafetyBench,MEDIC,,False,False,200,200,900,"test (MedSafety-Eval): 9 AMA-ethics categories x {GPT-4-generated, Llama-2-jailbreak-generated}",dc5d88e4c0100deada5a96f065a302959e6523a7,MIT (LICENSE); README: research use only,https://raw.githubusercontent.com/AI4LIFE-GROUP/med-safety-bench/dc5d88e4c0100deada5a96f065a302959e6523a7/README.md; https://raw.githubusercontent.com/AI4LIFE-GROUP/med-safety-bench/dc5d88e4c0100deada5a96f065a302959e6523a7/LICENSE; https://raw.githubusercontent.com/AI4LIFE-GROUP/med-safety-bench/dc5d88e4c0100deada5a96f065a302959e6523a7/datasets/test/gpt4/med_safety_demonstrations_category_1.csv; and 17 more files listed in registry/manifests/medsafetybench.json,Item text as released: the harmful medical request.,0.396
MedXpertQA,Medmarks,Alibaba,False,True,200,200,2448,test of the Text subset (Text/test.jsonl),7e7c465a68eb2b866926bfa59c8c9d17a8daba65,mit,https://huggingface.co/datasets/TsinghuaC3I/MedXpertQA/resolve/7e7c465a68eb2b866926bfa59c8c9d17a8daba65/Text/test.jsonl; https://huggingface.co/datasets/TsinghuaC3I/MedXpertQA/resolve/7e7c465a68eb2b866926bfa59c8c9d17a8daba65/MM/test.jsonl; https://huggingface.co/datasets/TsinghuaC3I/MedXpertQA/resolve/7e7c465a68eb2b866926bfa59c8c9d17a8daba65/README.md,"multiple choice: question stem (inline 'Answer Choices:' block removed), then the source's lettered options A-J one per line.",0.382
MeQSum,CLUE; MedS-Bench,,False,False,200,200,1000,whole set (no official split),51e36cf65c12602333b6fdc3ae8ecf10430529f7,CC BY 4.0 (README),https://raw.githubusercontent.com/abachaa/MeQSum/51e36cf65c12602333b6fdc3ae8ecf10430529f7/README.md; https://raw.githubusercontent.com/abachaa/MeQSum/51e36cf65c12602333b6fdc3ae8ecf10430529f7/MeQSum_ACL2019_BenAbacha_Demner-Fushman.xlsx,"Instruction-style via the CLUE suite template (zero-shot): CLUE system prompt, blank line, then the user turn wrapping the consumer health question (CHQ).",0.376
MTS-Dialog,BRIDGE,,True,False,200,200,597,test,41159677fba2454e032574625e24b24d29a17684,see BRIDGE-Open dataset card and the original dataset licence,https://huggingface.co/datasets/YLab-Open/BRIDGE-Open/resolve/41159677fba2454e032574625e24b24d29a17684/Dataset/MEDIQA 2023-chat-A.SFT.json; https://huggingface.co/datasets/YLab-Open/BRIDGE-Open/resolve/41159677fba2454e032574625e24b24d29a17684/Dataset/MEDIQA 2023-sum-A.SFT.json; https://huggingface.co/datasets/YLab-Open/BRIDGE-Open/resolve/41159677fba2454e032574625e24b24d29a17684/Dataset/MEDIQA 2023-sum-B.SFT.json,"instruction-style: the BRIDGE instruction, a blank line, then the input.",0.567
MTSamples,BRIDGE; MedHELM; Medmarks,,True,False,200,200,1036,whole set (test only): mtsamples_processed (treatment-plan task) + mtsample_procedure (procedures task),ebc104a4f96c5b7602242f301e081e9934a23344,"none stated (repo README: ""Benchmark data source for testing MedHELM""); notes originate from mtsamples.com",https://raw.githubusercontent.com/raulista1997/benchmarkdata/ebc104a4f96c5b7602242f301e081e9934a23344/README.md; https://raw.githubusercontent.com/raulista1997/benchmarkdata/ebc104a4f96c5b7602242f301e081e9934a23344/mtsamples_processed/1-year-old%20Exam%20-%20H%26P.txt; https://raw.githubusercontent.com/raulista1997/benchmarkdata/ebc104a4f96c5b7602242f301e081e9934a23344/mtsamples_processed/2-D%20Doppler.txt; and 1696 more files listed in registry/manifests/mtsamples.json,"Pooled components. MedHELM items: full mtsamples_replicate prompt exactly as HELM builds it (instruction, blank line, note with PLAN section removed, newline, final cue ""Answer:""; notes without a PLAN/SUMMARY/FINDINGS line skipped as in HELM). BRIDGE MTS and MTS-Temporal: instruction, blank line, input.",0.278
Pharmacology QA for Emerging Drugs,ClinicBench,,True,False,200,200,213,"whole set (sheet ""selected""; no split)",57138be8a73f09bef8980f849789baf57570c5cf,Apache-2.0 (ClinicBench repository LICENSE),https://raw.githubusercontent.com/AI-in-Health/ClinicBench/57138be8a73f09bef8980f849789baf57570c5cf/Pharmacology-QA.xlsx; https://raw.githubusercontent.com/AI-in-Health/ClinicBench/57138be8a73f09bef8980f849789baf57570c5cf/README.md; https://raw.githubusercontent.com/AI-in-Health/ClinicBench/57138be8a73f09bef8980f849789baf57570c5cf/LICENSE,"Multiple choice: question, then options A-D one per line.",0.304
PMC-Patients basic information extraction,MedS-Bench,,False,False,200,200,16704,"test (MedS-Bench test file, all instances)",6945b14e0f408082aecb6a77a185b9ebdd5fb413,MedS-Bench card states no license; PMC-Patients is tagged cc-by-nc-sa-4.0 on the Hub.,https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/Information_extraction/task74_pmc_patient_case_report_basic_information_extraction.json; https://huggingface.co/datasets/Henrychur/MedS-Bench-SPLIT/resolve/bae9b49bb23ffcd5b1b64345f021d4d5b803b503/Information_extraction/task74_pmc_patient_case_report_basic_information_extraction.json; https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/README.md,"instruction-style: the MedS-Bench task Definition (instruction), a blank line, then the instance input.",0.381
PUBHEALTH,MedS-Bench,,False,False,200,200,2423,"test (MedS-Bench test file, all instances)",6945b14e0f408082aecb6a77a185b9ebdd5fb413,MedS-Bench card states no license; health_fact is tagged MIT on the Hub.,https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/Fact_verication/task16_test_healthfact_classification.json; https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/Explanation/task18_test_healthfact_sentence_generation.json; https://huggingface.co/datasets/Henrychur/MedS-Bench-SPLIT/resolve/bae9b49bb23ffcd5b1b64345f021d4d5b803b503/Fact_verication/task16_test_healthfact_classification.json; and 2 more files listed in registry/manifests/pubhealth.json,"instruction-style: the MedS-Bench task Definition (instruction), a blank line, then the instance input.",0.206
PubHealthBench,Medmarks,,False,False,200,200,7928,test (PubHealthBench-Full); reviewed subset also downloaded,3c2d456075ba465a84f94eb1ffc8804e9a655842,cc-by-4.0 (contains UK public-sector information under the Open Government Licence v3.0),https://huggingface.co/datasets/Joshua-Harris/PubHealthBench/resolve/3c2d456075ba465a84f94eb1ffc8804e9a655842/data/test-00000-of-00001.parquet; https://huggingface.co/datasets/Joshua-Harris/PubHealthBench/resolve/3c2d456075ba465a84f94eb1ffc8804e9a655842/data/reviewed-00000-of-00001.parquet; https://huggingface.co/datasets/Joshua-Harris/PubHealthBench/resolve/3c2d456075ba465a84f94eb1ffc8804e9a655842/README.md,"multiple choice: question, then the seven source-lettered options (bench_options_list_formatted, 'A. ...') one per line.",0.388
PubMedQA,ClinicBench; MEDIC; MIRAGE; MedHELM; MedS-Bench; Medmarks; MultiMedQA,Anthropic; Google,True,True,200,200,500,PQA-L test (500 PMIDs listed in test_ground_truth.json),1cbae8e92f72f20c8d3747cbb3bf5bc53554d997,MIT (LICENSE),https://raw.githubusercontent.com/pubmedqa/pubmedqa/1cbae8e92f72f20c8d3747cbb3bf5bc53554d997/README.md; https://raw.githubusercontent.com/pubmedqa/pubmedqa/1cbae8e92f72f20c8d3747cbb3bf5bc53554d997/LICENSE; https://raw.githubusercontent.com/pubmedqa/pubmedqa/1cbae8e92f72f20c8d3747cbb3bf5bc53554d997/data/ori_pqal.json; and 2 more files listed in registry/manifests/pubmedqa.json,"Vignette + question: abstract contexts (one ""Label. text"" line each, as in Lievin et al./MedHELM), a blank line, then the question; no options are released (answers are yes/no/maybe).",0.426
RaceBias,MedHELM,,True,False,167,167,167,test (HELM uses all items as test),npj Digital Medicine ESM MOESM1; HELM 63754d05db6f874e41a395880fb573890a13e791,CC BY 4.0 (npj Digital Medicine open-access article),https://static-content.springer.com/esm/art%3A10.1038%2Fs41746-023-00939-z/MediaObjects/41746_2023_939_MOESM1_ESM.docx; https://raw.githubusercontent.com/stanford-crfm/helm/63754d05db6f874e41a395880fb573890a13e791/src/helm/benchmark/scenarios/race_based_med_scenario.py,"template: MedHELM race_based_med prompt (scenario text ""For the following question-answer pair ... Question: ... Response: ..."" inside the HELM multiple-choice-joint adapter with options A. yes / B. no and the output cue)",0.254
RCT-Text,MedS-Bench,,False,False,100,100,100,MedS-Bench-SPLIT test file (not in the full MedS-Bench release),HF Henrychur/MedS-Bench 6945b14e0f408082aecb6a77a185b9ebdd5fb413; HF Henrychur/MedS-Bench-SPLIT bae9b49bb23ffcd5b1b64345f021d4d5b803b503 (GitHub MAGIC-AI4Med/MedS-Ins 185cfc04a27023983e48b357a6f2fd525049b9af),"MedS-Ins/MedS-Bench GitHub LICENSE: ""CC BY-SA"" (MAGIC consortium); HF dataset cards declare no license; task-file Instance License: none given; source ""RCT"" via huggingface.co/datasets/allenai/mslr2022 (Cochrane subset; Wallace et al. 2020, arXiv 2008.11293)",https://huggingface.co/datasets/Henrychur/MedS-Bench-SPLIT/resolve/bae9b49bb23ffcd5b1b64345f021d4d5b803b503/Text_summarization/task110_rct_text_summurization.json,"instruction-style (MedS-Bench): the task Definition, a blank line, then the instance input (few-shot examples and outputs omitted)",0.206
SCT-Bench,MAST; Medmarks,,False,False,174,174,174,public subset (Adelaide SCT + Open Medical SCT); remaining 576 of 750 held out,c2843137b36218eac3f47e76eaa559c9fe7973e9,MIT (LICENSE),https://raw.githubusercontent.com/SCT-Bench/sctpublic/c2843137b36218eac3f47e76eaa559c9fe7973e9/README.md; https://raw.githubusercontent.com/SCT-Bench/sctpublic/c2843137b36218eac3f47e76eaa559c9fe7973e9/LICENSE; https://raw.githubusercontent.com/SCT-Bench/sctpublic/c2843137b36218eac3f47e76eaa559c9fe7973e9/data/sct_cleaned_full.csv; and 2 more files listed in registry/manifests/sct_bench.json,"Repository prompt template filled exactly (zero-shot, with-reason variant): guideline.md + testcase.md with scenario, hypothesis and additional information.",0.168
Species-800,MedS-Bench,,False,False,200,200,1539,"MedS-Bench test file(s); MedS-Bench-SPLIT has 1,500 (random subset)",HF Henrychur/MedS-Bench 6945b14e0f408082aecb6a77a185b9ebdd5fb413; HF Henrychur/MedS-Bench-SPLIT bae9b49bb23ffcd5b1b64345f021d4d5b803b503 (GitHub MAGIC-AI4Med/MedS-Ins 185cfc04a27023983e48b357a6f2fd525049b9af),"MedS-Ins/MedS-Bench GitHub LICENSE: ""CC BY-SA"" (MAGIC consortium); HF dataset cards declare no license; task-file Instance License: none given; source huggingface.co/datasets/species_800",https://huggingface.co/datasets/Henrychur/MedS-Bench/resolve/6945b14e0f408082aecb6a77a185b9ebdd5fb413/NER/task128_test_species800_named_enetity_recognition.json; https://huggingface.co/datasets/Henrychur/MedS-Bench-SPLIT/resolve/bae9b49bb23ffcd5b1b64345f021d4d5b803b503/NER/task128_test_species800_named_enetity_recognition.json,"instruction-style (MedS-Bench): the task Definition, a blank line, then the instance input (few-shot examples and outputs omitted)",0.206
USMLE practice questions,,Anthropic,False,True,200,200,325,"all sample items (Step 1 119, Step 2 CK 120, Step 3 137)",booklet versions: Step 1 updated April 2026; Step 2 CK updated July 2023; Step 3 updated June 2026 (files served under /sites/default/files/2021-10/),"Copyright FSMB and NBME, all rights reserved; booklets marked ""For Public Release""",https://www.usmle.org/sites/default/files/2021-10/Step_1_Sample_Items.pdf; https://www.usmle.org/sites/default/files/2021-10/Step2_CK_Sample_Questions.pdf; https://www.usmle.org/sites/default/files/2021-10/Step3_Sample_Items.pdf,"multiple choice: the question stem (preceded by the shared abstract or vignette, then a blank line, for items in a set) then the options one per line as ""A. ..."" (source letters)",0.316