Healthcare AI Benchmarks: Current Frontier Results

Index updated August 16, 2026 · 17 benchmarks · 70 models

This index tracks every healthcare benchmark on which current frontier models hold a public result: 17 boards across 17 sources as of August 16, 2026, each score carrying the source it came from. No single model owns the field, though Claude Opus 5 leads more than one board. The table links to a page per benchmark and a page per model.

The benchmarks

One board per benchmark, bars scaled to that board's own leader. Scores sit on different scales and never compare across boards.

Scores appear exactly as each source publishes them. Two of the boards, HealthBench Professional and HealthBench Hard, are full leaderboards Arcophos runs.

How to read this index

Healthcare evaluation splintered as the frontier moved: rubric benchmarks, hard subsets, agentic clinical simulations, and coding or documentation tasks each measure something the others miss, and their scores live on incompatible scales. This index keeps them side by side without pretending they add up. For any benchmark, its page carries what it measures and where its numbers come from; for any model, its page collects every board it appears on. The sourcing rules are on the methodology page.

Models across boards

15 of the 70 tracked models hold results on more than one board. Each has a page gathering everything the index knows about it, listed on the models page.

What is not here

Some well-known names are absent on purpose. A benchmark leaves the index when frontier models stop being tested on it, and these are the notable cases with the reason each one dropped out:

Data

The index is downloadable in full as JSON and CSV, CC BY 4.0, from fixed paths. Citation guidance and version history live on the data page.