Why benchmarks
What a benchmark tests is what gets fixed.
AI companies are graded on public benchmarks and compete to raise their scores. Benchmarks today test exam knowledge, not whether an answer connects the mouth to the rest of the body, or whether it helps someone actually get care. Dental records have the same gap: the diagnosis is written down, in pieces, and nobody checks whether the pieces agree.
Accurate AI answers
Score the AI that millions of people ask on the questions that matter at the seam of medical and dental care: medicines, chronic disease, emergencies.
Trustworthy records
Measure the data quality of electronic dental records, so a diagnosis can be counted, followed over time and linked to diabetes care.
Open and public
Questions trace to published evidence, scores are public, and the code, the synthetic data and the methods are open for anyone to check.
Benchmark 1 · dental records
PerioBench
The diagnosis is in the dental record. Nobody can count it.
Gum disease (periodontitis) is written down at almost every dental visit, but in pieces: notes, forms, codes and measurements that often disagree. So it is missed, can't be followed over time, and can't be linked to diabetes care. PerioBench scores how accurately rules and AI diagnose it from the whole record. It runs end to end on synthetic records:
- the whole record, combined by rules that read the gum measurements first: 234 of 240 charts right;
- the free-text note alone, read by a small trained model: 219 of 240;
- a bigger medical AI used as is: 185 of 240.
Scored against the generator's truth on synthetic charts, not clinical performance. Read in a different order, the same evidence gets 173: the order is one of the findings.
Benchmark 2 · AI answers
OralHealthBench
AI passes the dental exam. It fails the real patient.
Twelve leading AI models average 81% on the dental licensing exam but 22% on real patient cases (GlobalDentBench, 2026). No public test we found checks whether they connect the mouth to the rest of the body: a medicine that makes a tooth extraction unsafe, or gum disease and diabetes. OralHealthBench puts those questions, each traced to published evidence, on a public scorecard, and measures how much an open evidence agent improves the answers.
Its companion, AccessBench, asks whether an answer works for someone with no coverage, no car or no dentist nearby.
How the two connect
One harness, from gum charts to the questions people ask.
The dental record path builds synthetic patients whose right answers are known, in OMOP, the data format medicine already uses. Those patients become test questions for the lab. The same models, runners, harm grades and leaderboards score AI on each step.
Synthetic patients
Records with known right answers, in OMOP. No real patient data, ever.
Gum charts
PerioBench: can rules and AI read the whole dental record accurately?
Case questions
The same patients become questions, scored for accuracy and harm.
Medical-dental questions
OralHealthBench and AccessBench: the questions people actually ask.