One manifest lists every openFDA bulk file and the size_mb figure FDA publishes for each part.
FDA public data as JSON APIs and bulk downloads: drug labels, recalls, device reports, NDC listings, and drug adverse events. Adverse events are their own card because that download is a different decision.
The front door. The adverse-event warehouse behind it is the FAERS card.
No single file. Drug adverse events alone list 110,752.84 MB across 1,767 parts (see the FAERS card).
Formats
JSON API · bulk JSON (zipped)
License
Public domain (CC0)
Updated
Manifest last_updated 2026-10-08
How to download
Open the downloads page and pick one endpoint (labels, NDC, enforcement, or events).
For adverse events, use the FAERS card. That archive is the large one.
Each part is a zip of JSON. Download the parts you need; do not mirror them here.
The complete public download of U.S. drug adverse-event reports, file by file.
Quarterly bulk zips of drug adverse-event reports: the public FAERS ledger as JSON, from 2004 through the newest quarter in the manifest, plus an “all other” bucket.
This is the download people mean when they say “just grab FAERS.”
110,752.84 MB compressed across 1,767 files (sum of openFDA size_mb, export 2026-10-05).
111 GB compressed. about 2 hours on a 100 Mbps connection (estimate).
Unzipping needs more disk than the compressed file. Measured 2026-10-08.
The structured text of U.S. prescription labels, as the label XML itself.
The current FDA label (Structured Product Labeling) for marketed U.S. drugs — full prescribing information as structured XML, in bulk ZIP releases.
The fine print of every pill bottle in the country, machine-readable.
The DailyMed bulk page did not publish one combined byte count on the read dated 2026-10-08. openFDA’s separate drug-label JSON lists 1,773.39 MB across 14 files; that is a different packaging, not this XML.
The official therapeutic-equivalence list pharmacists use for generic substitution.
Approved drug products with therapeutic-equivalence evaluations — the list behind generic substitution, plus patent and exclusivity data.
The book pharmacists argue with. Now a ZIP file.
2.34 MB for openFDA’s Orange Book JSON zip (size_mb, export 2026-10-08). FDA’s own delimited zip is linked from the Orange Book data-files page; that older media URL returned 404 on this probe, so its byte count is not recorded.
2.3 MB compressed. under a minute on a 100 Mbps connection (estimate).
Unzipping needs more disk than the compressed file. Measured 2026-10-08.
Formats
JSON zip (openFDA) · TXT (delimited) from FDA
License
Public domain
Updated
Monthly (openFDA export 2026-10-08)
How to download
For a small complete JSON copy, download the openFDA zip in the command below.
For FDA’s delimited text, use the Orange Book data-files page and take the zip linked there.
This command fetches the official file named in the size line.
A CC0 identifier crosswalk for drugs. The full DrugBank database is a different, licensed product.
The freely licensed subset of DrugBank: the open drug and drug-target vocabulary (identifiers, names, CAS/UNII mappings) under CC0. The full database requires a license.
Byte size was not visible without signing in. The releases page is behind a free account, so this catalog does not guess.
Formats
CSV (zipped)
License
CC0 1.0 (open subset only)
Updated
With DrugBank releases
How to download
Create a free DrugBank account.
Open the latest release and download only the files marked Open Data (CC0).
Leave the licensed full database alone unless you have that license.
The identifier map that lets pharmacy systems and hospital systems name the same drug.
The normalized naming system for U.S. drugs — clinical drug, dose form, and brand mapped to stable identifiers. Full monthly releases need a free UMLS account.
The Rosetta Stone for why your pharmacy’s computer and your hospital’s computer disagree.
The full-file byte count sits behind the UMLS login. It is not recorded here.
Formats
RRF (pipe-delimited) · API (JSON)
License
Free; UMLS Metathesaurus License (free account) for full files
Updated
Monthly
How to download
Request a free UMLS license account if you do not have one.
From the RxNorm files page, download the current monthly release.
The RxNorm API can answer one drug at a time without the full file.
The exam-and-lab sample behind national estimates of cholesterol, blood pressure, and similar measures.
The National Health and Nutrition Examination Survey — interviews, exams, and lab measurements on a U.S. sample, in cycles since 1999. Each cycle is many component files.
The reason we know what the average American’s cholesterol actually is.
No combined size is published for every component. Demographics for the August 2021–August 2023 cycle (DEMO_L.xpt) is 2,582,160 bytes (Content-Length).
County, city, and tract estimates for measures that a national survey cannot publish that small.
Model-based estimates of health measures for counties, places, and census tracts — the local cut of measures that national surveys cannot print at that geography.
PLACES is many tables. A single byte count for “all of PLACES” was not recorded, so none is shown.
Formats
CSV via data.cdc.gov
License
Public domain
Updated
Annual releases
How to download
Open the PLACES data portal and choose one geography (county, place, or tract) and one year.
Export that table as CSV from data.cdc.gov.
Read the measure definitions before comparing two tracts.
The shelf of Medicare payment and provider files. This card is the catalog, not one file.
The catalog of CMS program datasets: Medicare utilization and payment, hospital cost reports, provider enrollment, and more. Each dataset is its own download.
No single file. Each dataset on data.cms.gov has its own size, shown on that dataset’s page.
Formats
CSV · API (JSON)
License
Public domain
Updated
Dataset-dependent
How to download
Search for one dataset (for example Medicare physician utilization).
Read that dataset’s page for the file size CMS prints there.
Download that file, or use its API, rather than trying to take the whole catalog.
The public survey of what U.S. households actually spent on health care.
Medical Expenditure Panel Survey — nationally representative U.S. data on health service use, cost, and insurance. AHRQ publishes each year’s files separately.
Each PU file has its own size on the download page. A total for every MEPS file was not summed here.
Formats
ASCII, SAS, Stata, or XLSX by file
License
Public domain
Updated
Annual files
How to download
Choose a data year and a file (full-year consolidated is the usual start).
Download the format you can read (SAS transport or ASCII plus the codebook).
Check the file’s own size on that row before you start.
A synthetic Medicare claims set you can download without a data-use agreement.
Medicare Claims Synthetic Public Use Files (2008–2010): synthetic beneficiary, carrier, inpatient, outpatient, and prescription claims. Synthetic means the rows are not real people.
CMS publishes 20 samples. A byte total was not recorded on this probe, so none is shown. These files are synthetic.
Formats
CSV zip, by sample
License
Public domain (synthetic)
Updated
Fixed 2008–2010 release
How to download
Read the page so you know the files are synthetic, not a 5% real-claims extract.
Download one sample (Sample 1 is enough to learn the layout).
Beneficiary, carrier, inpatient, outpatient, and PDE files are separate zips inside a sample.
The bulk XML of PubMed’s annual baseline, file by file.
The annual PubMed citation baseline — biomedical citations and abstracts with MeSH indexing — as many gzipped XML files, plus daily update files outside this folder.
All of biomedicine’s abstracts in one download. Bring a real disk.
1,334 baseline files (pubmed26n0001 through pubmed26n1334). Adding the FTP listing’s rounded megabyte sizes comes to about 52 GB. Each line is rounded, so the total is approximate.
52 GB compressed. about 69 minutes on a 100 Mbps connection (estimate).
Unzipping needs more disk than the compressed file. Measured 2026-10-08.
The credentialed ICU record set most clinical machine-learning papers mean by “MIMIC.”
De-identified electronic health records from Beth Israel Deaconess ICU and ED stays. Version 3.1 is the current credentialed release. The open demo is a separate card.
Free, but you must pass a human-subjects course first. The data is worth the homework.
PhysioNet did not print a file size on the public version page read on 2026-10-08 (the bytes sit behind the data-use agreement). The open demo’s uncompressed size is on the demo card.
Formats
CSV · BigQuery
License
PhysioNet Credentialed Health Data License (free; DUA + CITI training required)
Updated
Version 3.1 is the current release listed on the project page
How to download
Complete the required human-subjects training and upload the completion report.
Sign the data-use agreement on the version 3.1 page.
Only then does PhysioNet show the files. Use the MIMIC-IV Demo card if you want the schema with no agreement.
An open demo of multi-hospital ICU stays, separate from single-center MIMIC.
A demo of the multi-center eICU database: de-identified ICU stays from many hospitals, open without the credentialed agreement. The full eICU set is credentialed and is not this card.
PhysioNet prints “Total uncompressed size: 130.6 MB.” Compressed size was not listed, so download time is not estimated.
Formats
CSV
License
Open access (demo); see the PhysioNet page
Updated
Demo version 2.0.1
How to download
Open the demo page and download the files listed there.
The full eICU database is a different, credentialed project. This card is only the demo.
130.6 MB is the uncompressed total PhysioNet prints.
Synthetic patients you generate yourself, with no real medical records in the output.
A synthetic patient generator. It writes fictional FHIR or CSV records. There is no one official patient file — you choose how many patients to generate.
No fixed dataset file. Output size depends on how many patients you ask the generator to write.
Formats
FHIR · CSV (generated)
License
Apache-2.0
Updated
Generator releases on GitHub
How to download
Read the README and install the generator from the GitHub release.
Generate a small population first (for example 100 patients) and look at the CSV or FHIR.
Scale the run only after that sample looks like what you wanted. The records are fictional.
The public compound identifier and title table. Not the full 3D structure dump.
PubChem’s compound identifier tables. This card sizes the CID–title file. The full structure dump is a different, larger FTP tree and was not summed.
1.8 GB for CID-Title.gz (FTP listing, rounded, dated 2026-09-27). The full compound SDF collection is a separate directory and is not included in that figure.
1.8 GB compressed. about 2 minutes on a 100 Mbps connection (estimate).
Unzipping needs more disk than the compressed file. Measured 2026-10-08.
Formats
gzipped TSV
License
Public domain (U.S. Government work)
Updated
Listing dated 2026-09-27
How to download
Download CID-Title.gz if you need compound IDs and names.
Open the Extras directory for the other identifier tables (SMILES, InChI-Key, synonyms). Their listing sizes are on that page.
Do not treat this file as the complete structure dump.
This command fetches the official file named in the size line.
This library is a personal, hand-maintained list of public resources.
Each dataset belongs to and is licensed by its maintainer — check the
license before you build on one. Listing here is not an endorsement by
any of these organizations, and nothing on this page is medical advice.
Byte counts come only from the maintainer’s own page, manifest, or
directory listing. A blank size means they did not publish one.