Open data

Name: Clinical Benchmarks results
Creator: Clinical Benchmarks
License: https://creativecommons.org/licenses/by/4.0/

Every result on this site as files and static JSON, current to 30 September 2026 (revision 7165).

Downloads

Results CSVOne row per result, with rank on its own board, source link, locator and quote./data/results.csv
Benchmarks CSVOne row per board: publisher, scale, direction, links./data/benchmarks.csv
Models CSVEvery model with a result here: lab, kind, release date, list price when known./data/models.csv
Sources CSVEvery document cited, with kind, publisher and dates./data/sources.csv
Index CSVThe Clinical Benchmarks Index: each ranked model, its score with both parts (mean placement and confidence multiplier) and its result on each board./data/index.csv
Everything JSONBenchmarks, results, models, sources and the change feed in one file./data/benchmarks.json

JSON endpoints

Plain static files: no key, no rate limit beyond the host's, CORS open. Each response wraps its records in the same envelope: dataset name, revision, last update, licence and the list under data.

Endpoint	Returns
/api/v1/benchmarks.json	All boards with their facts, secondary measures and counts.
/api/v1/results.json	All results, headline and secondary, with evidence fields.
/api/v1/models.json	Models that have at least one result here.
/api/v1/sources.json	Documents behind the results.
/api/v1/index.json	The Clinical Benchmarks Index, this site's own calculation: method, boards used and every entry with its components.
/api/v1/benchmarks/healthbench-professional.json	One board and its rows. Replace the id with any board id.

Fields worth knowing

value is the number on the board's own scale, for sorting and plotting. valueLabel is how this site prints it. display is how the source printed it.
rankOnBoard and rowsOnBoard count rows on one board and one measure. They are not comparable across boards.
variant holds the configuration caveat (effort setting, harness, grader). Read it before quoting a row.
reportedBy is vendor, benchmark_owner (official leaderboard) or third_party (independent run). confidence is verified, partial, unverified or disputed.

Licence and citation

The compilation is released under CC BY 4.0. Credit Clinical Benchmarks and link to clinicalbenchmarks.ai. The system cards, papers and leaderboards the numbers come from keep their own terms.

Suggested citation: Clinical Benchmarks, revision 7165, 30 September 2026. https://clinicalbenchmarks.ai/data.

What changed between revisions is on the updates page. Found a wrong number? Tell us and name the document and page.