Open Data

Every record in the directory as JSON and CSV, free to reuse with attribution. Our proposed analysis is left out until it is approved.

Downloads

organisationsDevelopers, labs, vendors and publishers768organisations.json organisations.csv
modelsModel records with class, openness, size and where they run3,188models.json models.csv
edgesLineage: parent and child links between models504edges.json edges.csv
licencesLicence terms and commercial use75licences.json licences.csv
benchmarksBenchmarks and what they measure271benchmarks.json benchmarks.csv
resultsBenchmark scores, each with who ran it1,196results.json results.csv
runtimesInference runtimes and formats30runtimes.json runtimes.csv
registriesModel registries and catalogues36registries.json registries.csv
topicsR&D topics152topics.json topics.csv
usecasesIndustry use cases201usecases.json usecases.csv
deploymentsDocumented deployments by named organisations299deployments.json deployments.csv
model topicWhich models each R&D topic links4,230model_topic.json model_topic.csv
model usecaseWhich models each use case names182model_usecase.json model_usecase.csv

Where this comes from

Built from the same records as every page, at the same build. Lists inside a CSV cell are joined with a semicolon; nested objects are written as JSON. Proposed tier grades, their rationale and confidence are removed from topics until approved; opportunities, use case to topic links and use case ideas are not published.