Four-Factor Integrity Breakdown & Primary-Source Evidence
Access Model: Public Plaintext
"We use the HealthBench and HealthBench Professional datasets that OpenAI publishes with simple-evals (commit 652c89d), pinned by checksum: 5,000 and 525 conversations."
Originality
Weight 25%
10/10
Reuse Profile: All Original (0% reuse)
"Each example includes a physician-authored conversation with ChatGPT for Clinicians and is scored via rubrics written and iteratively adjudicated by three or more physicians across three phases."
Freshness
Weight 20%
3.5/10
Initial Release: 2025-05Latest Refresh: 2026-04Rolling Updates: No
"arXiv:2604.27470 (cs) [Submitted on 30 Apr 2026]"
Grading
Weight 15%
7.3/10
Integrity: Independent StandardizedSandbox: not applicable single turn
"Each model receives the conversation as user and assistant turns, with no system prompt and no tools, and writes the next assistant reply. A judge model checks the reply against each rubric item separately with simple-evals' grader prompt"
Frontier Model Standings on HealthBench
Top 15 active models tracked
Benchmark maintainers can embed this live SVG badge in GitHub READMEs or documentation:
[](https://benchaudit.org/benchmarks/healthbench/)