Medical Bench

About / medicalbench.ai

Count the images. Inspect the questions.

A medical VQA result depends on which images, question forms and answer rules were evaluated. We examine VQA-RAD’s clinician-generated questions and SLAKE’s visual and knowledge-based tasks, keeping paper counts separate from cleaned releases. Our original coverage map and guides explain what those choices support. Selected historical results are credited to the original papers; no new model runs are claimed.

Who this is for

This resource is written for research teams choosing task formats for medical AI benchmarks. Its scope is benchmark-specific analysis: what the published tasks measure, how scores are produced, and what the evidence can support.

  • Counting units. Distinguish images, question records and related question forms.
  • Task evidence. Separate image understanding from external-knowledge access.
  • Comparison policy. Keep release, language, split and answer judging visible.

Ownership and independence

Medical Bench is published by Arcophos. Arcophos also publishes other healthcare AI evaluation resources linked from this site. Those links are disclosed as related resources, not independent endorsements.

The site is not the official home of any third-party benchmark, regulator, hospital, or model developer discussed in its guides. Benchmark names and publication titles identify their respective authors’ work. Our checklists and interpretations are editorial material.

How to read this site

Start with a benchmark dossier, follow its original references, and use the analytical explorer to inspect the tasks or scoring assumptions. The planning worksheet remains available as a secondary tool. Each guide states a direct answer, develops its reasoning, and links supporting references.

These materials do not establish clinical efficacy, patient benefit, or regulatory compliance for any system. Our editorial method explains the boundaries.

Your worksheet data

Worksheet selections are stored in this browser on this device. They are not sent to Arcophos. You can reset selections at any time or download them as a text file. This site has no analytics or advertising scripts.

Contact and corrections

For a correction, include the page URL, the specific claim, and a link to the original evidence. Contact Arcophos ↗