Arcophos

The index is calculated from published results alone: a model’s weighted average over the boards it has results on, times a confidence multiplier that is full from six boards. Each column of dots is one point out of 100. The pips are the ten boards, numbered 01 to 10; point at a board to light its pips.Each row of dots in a column is one point out of 100. The ten boards are numbered 01 to 10; point at a board to see which models have a result on it.

All 150 models Method

What we do

Care

We build hospitals and clinics, and we operate multiple of them. Each one is a Clinical Deployment Environment [CDE], where care runs as a relay from healthcare administration to patient follow-up. Care comes first; the data follows.

How the clinics run

Research

We measure how frontier models perform on clinical work. Our Clinical Benchmarks Index, at clinicalbenchmarks.ai, reads published results from ten clinical benchmarks on one scale, each linked to the document it came from.

Our research

Systems

Each visit is designed to be captured as one episode. At four checkpoints the clinician commits a ranked differential before seeing what comes next; results stay sealed until the expectation is committed, and follow-up establishes the outcome.

About the data

Clinical Deployment Environment [CDE]

At the clinic

Healthcare Administration
Patient Triage
Examination
Lab Tests
Medication Prescription
Patient Follow-Up

EpisodeOne visit · dated

For research

Benchmarks
Datasets
RL Environments
Safety Evaluations
Research Papers

Each visit is designed to become an episode: one dated record of what was expected, what was found and what happened next. De-identified episodes are used to train and evaluate models.

How a visit runs

Recent work

Health Optimization BenchLeaderboard16 models · 257 tasks

Health Optimization Bench

healthoptimizationbench.com

16 frontier models on 257 rubric-graded tasks, across eight subject areas of preventive and optimization medicine.

Clinical Reasoning BenchIn development

Clinical Reasoning Bench

Research · Soon

Our second benchmark, now in development.

Multilingual clinical speechEast Africa

Multilingual clinical speech

Data · Overview

An overview of multilingual clinical speech from East Africa.

All work

How we work

In the clinic

Care runs as a relay, from healthcare administration and patient triage to medication prescription and patient follow-up. At four checkpoints the clinician commits a ranked differential before seeing what comes next, and results stay sealed until the call is made.

In the lab

We run controlled experiments in our Clinical Deployment Environments to measurably improve frontier models in long-horizon, non-verifiable healthcare domains. Our Clinical Benchmarks Index is public, and every number in it links to its source.

The frontier research lab for AI-native healthcare systems.

How AI models score on clinical work

Clinical Benchmarks Index