MODEL Data Pack

The data behind the MODEL page at https://indahl.ai/hermes/model, as files you can open in a spreadsheet or Python. MODEL sets what AI hardware vendors state about their systems next to the results MLCommons publishes. Each file comes as the page publishes it, and each JSON file also comes as CSV with one row per record. The statements were fetched and compared on 2026-10-28, the files were built on 2026-10-28, and the MLCommons files were retrieved on 28 October 2026. Vendor pages and MLCommons files change, so a later read can differ. Page: https://indahl.ai/hermes/model/data-pack How it was tested. On 1 November 2026 every CSV was opened with pandas 3.1.0 and the live MODEL page and its 74 source pages were fetched and compared. Every count matched, including 531 statements with 8 ABOVE, 51 BELOW, 368 CONSISTENT and 104 CANNOT_TELL, and 1,727 results by accelerator in 569 groups. One note on a Nebius storage row differed from the page, and the published file was corrected on 2 November 2026 and the pack carries the corrected file. Excel and LibreOffice were not tested, and the MLCommons files were not downloaded again. Terms. This pack mixes data from several sources and each part keeps the terms of its source. MLPerf results published by MLCommons Association are under the Apache License 2.0, and the text is in APACHE-2.0.txt in the zip. Vendor sentences are quoted from the sources named in each row. Heuremen LLC's own compilation, labels and text may be used with attribution to Heuremen LLC and https://indahl.ai/hermes/model. Full terms by file are in LICENSE.txt in the zip. The Zenodo licence field is set to Other (Attribution) because no single listed licence covers the whole pack.