Methodology
The numbers on this site are produced by running the published benchmark, not by collecting vendors' self-reported results. Every model faces the same task set under the same grader configuration, so a difference between two rows is a difference on identical work. What the benchmark itself measures and how it was constructed is on the benchmark page; this page covers the run protocol.
Run protocol
| Task set | The released 525-task HealthBench Professional set, unmodified. Task construction is OpenAI's and is described on the benchmark page. |
|---|---|
| Candidates | Each model is run through its vendor's public API under default sampling, one response per task, no tools, no retrieval, no system-prompt scaffolding. |
| Grader | The benchmark's default grader configuration: GPT-5.4, low reasoning effort. Criterion judgments are binary, and the published length adjustment is applied. |
| Score | Earned rubric points divided by possible points, clipped to 0 to 1, reported to three decimals. |
| Pricing and specs | Context windows and per-token prices come from vendor documentation and list price sheets, checked on the update date. Prices shown are base rates as of August 16, 2026. |
Comparability
Rubric scores are model-graded, so they are comparable only within an identical grader configuration. Numbers here can differ from OpenAI's paper by a few points: the paper's headline result for GPT-5.4, 0.590, was measured inside ChatGPT for Clinicians, with product scaffolding a raw API call does not have. Comparing a score from this table against a score graded elsewhere is not meaningful; comparing two rows of this table is.
Limitations
Single-sample scoring measures typical rather than best-of-k performance. The grader is itself a model, and grader bias is a known open problem for rubric benchmarks; using the benchmark's default grader keeps results reproducible, not infallible. Per-use-case subscores (care consult, documentation, research) are not yet broken out on this site. And a benchmark score is not a clinical safety certification: it measures rubric adherence on 525 hard tasks, nothing more.
Updates
The leaderboard is re-run when a frontier model ships or a vendor materially updates one. Every change, including price updates, is dated on the updates page, and each release of the results is published in full on the data page. The current table covers 9 models as of August 16, 2026.