Medical and clinical work
Clinical questions, visit notes, medical coding and chatbot safety with teens. The tests here score clinical answers on diagnosis, treatment and harmful advice, notes against expert checklists, codes against those chosen by certified coders, and whether chatbots keep to safety checks with teens.
Sources last checked 2026-10-23.
We can't show scores for this kind of work yet. These testers publish results: ARISE AI Research Network (one test) and Vals AI (three tests). See who tests it
At a glance
Vendor claims have not been collected for any category yet.
ⓘ More about this category
- Reliability not yet measured
- 4 sources, linked results only
- Sources last checked 2026-10-28
- Ring 4: needs rings 1 to 3 first
- MAST = MAST (Medical AI Superintelligence Test), MSC = MedScribe, MCD = MedCode, SAFE = SAFE-Teen
Tested by
MAST (Medical AI Superintelligence Test)
ARISE AI Research Network, checked 2026-10-28A combined score across clinical tests of diagnosis, treatment planning, avoiding harmful advice, radiology and medical images, built so weakness in any one area pulls the total down.
ARISE reports that some models reason well but underperform physicians on avoiding harm.
Last updated 2026-10-12Covers 22 of the models we trackScores not shown: permission requested on 2026-10-27
See their resultsMore about this test: MAST (Medical AI Superintelligence Test)
Reliability not measured by this source.
Technical details for MAST (Medical AI Superintelligence Test)
MAST General Composite: A combined score across clinical tests of diagnosis, treatment planning, avoiding harmful advice, radiology and medical images, built so weakness in any one area pulls the total down.
License: Score reuse not stated. Checked 2026-10-28.
MedScribe
Vals AI, checked 2026-10-28How well a model writes a doctor's visit note in the standard four-part format (what the patient reports, what the doctor finds, the assessment and the plan) from a visit transcript, scored against expert-written checklists.
Vals AI reports that most models do slightly worse on the plan section of the note.
Last updated 2026-10-13Covers 38 of the models we trackVals AI built this test's scoring checklists with Protege, which supplies real-world data for building AI.Scores not shown: permission requested on 2026-10-24
See their resultsMore about this test: MedScribe
Reliability not measured by this source.
Technical details for MedScribe
Accuracy: How well a model writes a doctor's visit note in the standard four-part format (what the patient reports, what the doctor finds, the assessment and the plan) from a visit transcript, scored against expert-written checklists.
License: Score reuse not stated. Checked 2026-10-28.
MedCode
Vals AI, checked 2026-10-28How well a model assigns standard diagnosis codes for a whole hospital stay from patient records, compared with codes chosen by two certified coders.
Vals AI reports that most errors happen when a model misses a diagnosis code entirely.
Last updated 2026-10-13Covers 38 of the models we trackVals AI built this test's set of patient records with Protege, which supplies real-world data for building AI.Scores not shown: permission requested on 2026-10-24
See their resultsMore about this test: MedCode
Reliability not measured by this source.
Technical details for MedCode
Accuracy: How well a model assigns standard diagnosis codes for a whole hospital stay from patient records, compared with codes chosen by two certified coders.
License: Score reuse not stated. Checked 2026-10-28.
SAFE-Teen
Vals AI, checked 2026-10-28How often a chatbot fails clinician-designed safety checks in long conversations with a simulated teen who hints at risk, asks for medical advice or seeks emotional attachment.
Vals AI reports that chatbots often failed to check teens' safety, gave diagnoses and medication advice too readily, and used language that builds emotional attachment.
Last updated 2026-10-13Covers 7 of the models we trackScores not shown: permission requested on 2026-10-24
See their resultsMore about this test: SAFE-Teen
Reliability not measured by this source.
Technical details for SAFE-Teen
Failure rate: How often a chatbot fails clinician-designed safety checks in long conversations with a simulated teen who hints at risk, asks for medical advice or seeks emotional attachment.
License: Score reuse not stated. Checked 2026-10-28.
What this doesn't tell you
- None of these tests used your patients, your records system or your clinical protocols.
- A good test score is not clinical validation. Medical use needs its own review.
- Scores from different sources are not comparable, even when they share a unit.
- None of these tests were run on your own documents, code or data.
- No source for this task allows us to show its figures yet. Follow the links for their results.
Data version 2026-10-28+91ac3fc9d071 · Terms of use