Skip to main content
Health-AI Watch · Evidence ledger

Recent models, without an invented ranking.

A searchable working record of release claims, outside observations, direct health evidence, policy behavior, access, cost, and unresolved questions.

← Health-AI Watch
How to use this ledger. Start with evidence stage, not model name. A release baseline records what a lab published; it is not independent evidence. A card matures only as stronger evidence arrives. No model has a Health-AI Watch score.

Find the model. Inspect what is actually known.

Reverse-chronological cards for major models a health researcher could plausibly use. Search, filter, or compare up to three without collapsing unlike evidence into one score.

Evidence cutoff July 20, 2026Unknown means not established by the cutoff—not zero, safe, or poor. Product routing and interface restrictions may differ from base-model capability.
2026 release trace
Loading the sourced catalog…
Loading releases…
Loading the sourced release catalog…
What this ledger does not do

It does not infer a health rank, treat evidence maturity as capability, or equate a base model with a consumer product. API price is not total benchmark cost, and an open-weight release still has hardware and deployment costs. The ledger identifies candidates and evidence gaps; it does not establish clinical readiness.

A release record chooses the candidate. It does not answer the question.

The main Watch asks whether selected systems can conduct a rigorous, reproducible health study—and records result fidelity, method, calibration, audit recovery, human burden, and cost.