Open data · reviewed 28 September 2026

Data and downloads

Name: Health Evals results index
Creator: Health Evals
License: https://creativecommons.org/licenses/by/4.0/

Download the data behind our MedPIC-Bench analysis: 28 published model rows, question counts, configuration context and source provenance. These files are generated from the same snapshot as the pages.

MedPIC-Bench

medpic.json ↓: All 28 model results, six coverage dimensions, department group results, denominators, source URLs, dataset revision and reproducibility limits.
medpic.csv ↓: One row per model, with six accuracy measures, the published GF–CF gap, denominators, configuration, source locator and dates.

Accuracy values use a 0–100 percentage scale. The GF–CF gap uses percentage points. The paper was published on 4 August 2026; the source review was on 28 September 2026. Unknown measurement dates remain null in JSON and empty in CSV. A source check is not an independent model run.

The JSON includes our counts of the public release. It does not contain the original question text, raw model responses or a reconstructed pair mapping. Download the original questions from the authors’ Hugging Face dataset. Read the methodology and limitations before combining or interpreting these measures.

Credit the benchmark and the analysis

Cite model scores to Zhitian Hou and colleagues, Evaluating Counterfactual Sensitivity to Patient Information in Medication-Safety Reasoning, arXiv:2608.03028v1, Table 2. For the department group means, cite Table 3.

Arcophos / Health Evals. MedPIC-Bench results and coverage analysis. Reviewed 28 September 2026. https://healthevals.com. Accessed [date].

The Arcophos compilation and analysis are available under CC BY 4.0. The original question dataset is also labeled CC BY 4.0 by its authors. This does not relicense the paper, model names or other upstream works.

Broader healthcare index

The 17-benchmark index remains available separately, with its existing URLs and export columns. Its snapshot is dated 2026-09-28; individual results retain their original measurement and source-review dates.

benchmarks.json: The broader index, source review status, citations and locators.
benchmarks.csv: One flat row per indexed result, with source, configuration and review status.
llms.txt: A compact map of the MedPIC analysis and wider index.
llms-full.txt: The MedPIC source rows and complete wider index in plain text.

Wider-index citation: Health Evals index, September 28, 2026 snapshot, https://healthevals.com/index. Index compilation: CC BY 4.0. Individual scores should be cited to their named sources. See the index methodology and review history.