Published by REAGEN, Inc.

What we do

We collect source-linked benchmark results and run our own model evaluations. HealthBenchAI team evaluations are labelled “Evaluated by HealthBenchAI” and link to individual articles with results, settings and supporting evidence. Scores from other evaluators remain attributed to their original sources. We do not sell rankings or accept payment for placement.

Our original model tests and later evidence checks are different activities. An arithmetic audit recomputes scores from retained records; it does not mean the team never ran the original test, nor does it constitute a new model run or independent regrading. Read our evaluation articles.

What we are not

HealthBench AI is an independent index. It is not affiliated with, endorsed by or operated on behalf of OpenAI, Google, Microsoft, the PubMedQA or MedQA authors, the USMLE program, or any model developer. The HealthBench benchmark name belongs to its owner; our use describes the data indexed. Nothing on this site is medical advice.

Principles

• Every score has a public source.
• Dates are shown, so age is visible.
• Overlaps and limits are stated plainly.
• Submissions are reviewed by a person before anything is published.

Questions or corrections go through the contact form.