Get started Console

Modern medicine75

Traditional medicine24

Cross-cutting12

Collections

Models8

License status and Mdexa hosting, recorded separatelyAll datasets

MedBench上海人工智能实验室

Verified by Mdexa · Checked against the source text under public rules.How we check

Commercial use
UnclearUnverified
Derivatives
UnclearUnverified
Redistribution
UnclearUnverified
File on Mdexa
NoPending

Data license not found

UnverifiablePendingNOASSERTIONSave

Badge

Mdexa verified · Unverifiable · 2026-11Mdexa verifiedUnverifiable · 2026-11

[![Mdexa verified · Unverifiable · 2026-11](https://mcp.mdexa.com/v1/badge/datasets/medbench.svg)](https://mdexa.com/datasets/medbench/)

Place it on a repository or dataset page; the status and date follow the verification. About badges

Cite this pageMdexa, MedBench（上海人工智能实验室）, https://mdexa.com/datasets/medbench/, verified 2026-11-12.

Key facts

Domain
Clinical medicine Modern medicine
Content
Question banks and benchmarksBenchmark (multi-specialty comprehensive evaluation)
Size
  • 300,901 questions (claimed)v1 (search snippet)
  • 43 categories (claimed)v1 clinical specialties
  • ≥700,000 tasksv4 (paper)
  • ≈1,000,000 questionsv5.0 (source material)
  • ≥70,000 questionsv5.0 open-sourced (source material)
As stated: v1：300,901题、43个临床专科（搜索摘要）；v4：70万+专家编制任务（论文称）；5.0 称近100万条试题、开源7万+条（来自素材）
Language / modality
Chinese / Text
Publisher
上海人工智能实验室 / OpenMedLab
Year
2024
Expert review
The paper says reviewed by clinicians from 500+ institutions (v4, from source material)
Verification
Unverified

Commercial use of MedBench（上海人工智能实验室）

Commercial use: Unclear (Unverified). License status: Unverifiable.

Quoted statements and their sources are in MedBench（上海人工智能实验室） license evidence.

Verdicts record what sources state and are not legal advice.

Tags

Intended use
Evaluation

Quoted from the sourceTitle: MedBench v4: A Robust and Scalable Benchmark for Evaluating Chinese Medical Language Models, Multimodal Models, and Intelligent AgentsSnapshot s01

Specialty
Multi-specialty

Quoted from the sourced-based benchmarking infrastructure comprising over 700,000 expert-curated tasks spanning 24 primary and 91 secondary specialties, with dedicated tracks for LLMs, multimodal models, and agents. Items undergo multi-stageSnapshot s01

Source type
Expert-written
Expert involvement
Reviewed, method unspecified

Author-statedInferredHuman-confirmed

MedBench（上海人工智能实验室） license evidence 2

Repositories returned 404 and the platform requires login

  1. Search result snippetSearch snippet only

    Search results showed no data license [Note: No URL recorded, so no snapshot exists]

    Second-hand, not verified

    No link—
  2. PaperSecond-hand only
    CC-BY-4.0

    SciOpen version of the paper is CC BY 4.0, covering the paper only [Note: No URL recorded, so no snapshot exists]

    Second-hand, not verified

    No link—

Source and lineage

The paper says expert-authored with clinician review

Known issues with MedBench（上海人工智能实验室）

  • The evaluation platform requires login to obtain data; github.com/open-compass/MedBench and OpenMEDLab/MedBench both returned 404 this time
  • Only part of the full set is open-sourced

Preflight

Not run Runs only on ingested data; says nothing about quality.

Downloading MedBench（上海人工智能实验室） and API access

No file on Mdexa (see the verdict). Get it from the source and confirm the terms there:

Citing MedBench（上海人工智能实验室）

Mdexa, MedBench（上海人工智能实验室）, https://mdexa.com/datasets/medbench/, verified 2026-11-12.
More citation details

The source gives no citation text; the line above cites this card.

Machine-readable

medbench.jsonmedbench.mdcatalog.json

Change log 6
2026-11-13Added specialty tags: multi-specialty(author-stated).
2026-11-13Removed the stored size-tier tag; the tier is now computed from the size field when pages are built.
2026-11-13Added dataset-level tags: 5 (1 stated by the authors with excerpts, 4 inferred from the content). No conclusion changed.
2026-11-12Added verbatim license excerpts and snapshot ids from captured page snapshots; no conclusion changed.
2026-11-12License status split into license_status (what the license says) and hosting (what Mdexa does); the earlier conclusion is kept for traceability.
2026-11-12Card converted from an earlier manual catalog; no new facts added; reclassified under taxonomy v2.

Listed in

Datasets similar to MedBench（上海人工智能实验室）

  • CMBChinese Medical BenchmarkVerified by MdexaMultiple-choice (single answer)EvaluationWith explanations PermissiveLink only 11.2K questions

    License is permissive, but source risk: questions come from copyrighted exams; link only.

    Clinical medicine · Question banks and benchmarks · Clinical cases · Exam-derived

    Open the cardSave

    Shared: Intended use Evaluation · Specialty Multi-specialty · domain Clinical medicine · content Question banks and benchmarks

  • DiagnosisArenaVerified by MdexaMultiple-choice (single answer)Evaluation RestrictedLink only —

    License restricted: restrictions are read broadly: forbids commercial use. Risk flag(s): Real patient text. Link only.

    Clinical medicine · Question banks and benchmarks · Clinical cases · Real patient text

    Open the cardSave

    Shared: Intended use Evaluation · Specialty Multi-specialty · domain Clinical medicine · content Question banks and benchmarks

  • MedMCQAVerified by MdexaMultiple-choice (single answer)Fine-tuningWith explanations PermissiveLink only —

    License permissive: first-hand data-scope statement(s) allow commercial use and redistribution. Risk flag(s): Exam-derived, scraped. Link only.

    Clinical medicine · Question banks and benchmarks · QA, dialogue and instruction data · Exam-derived · Scraped

    Open the cardSave

    Shared: Intended use Evaluation · Specialty Multi-specialty · domain Clinical medicine · content Question banks and benchmarks

  • MedQAUSMLE 4-option, US subset + English textbook corpusVerified by MdexaMultiple-choice (single answer)Evaluation PermissiveLink only —

    License permissive: first-hand data-scope statement(s) allow commercial use and redistribution. Risk flag(s): derived, Exam-derived. Link only.

    Clinical medicine · Question banks and benchmarks · Literature · Derived · Exam-derived

    Open the cardSave

    Shared: Source type Expert-written · Intended use Evaluation · Specialty Multi-specialty · domain Clinical medicine · content Question banks and benchmarks

  • CBLUEVerified by MdexaNamed entity recognitionEvaluation UnstatedPending 8 tasks

    License unstated: Only a code repository license; no data license stated; link only until resolved.

    Clinical medicine · Question banks and benchmarks · Annotated corpora

    Open the cardSave

    Shared: Intended use Evaluation · Specialty Multi-specialty · domain Clinical medicine · content Question banks and benchmarks