# HealthBench Contamination Audit

> HealthBench scores 4.3/10 (Tier C, High Risk) on BenchAudit as of November 2026, reflecting publicly downloadable plaintext test files and standardized independent third-party grading.

- **Canonical URL:** https://benchaudit.org/benchmarks/healthbench/
- **Benchmark ID:** `healthbench`
- **Hub:** Artificial Analysis / OpenAI
- **Category:** Science
- **Domains:** Medical
- **Skills:** Knowledge Work, Reasoning, Writing, Instruction Following
- **BenchAudit Score:** 4.3/10 (Tier C - High Risk)
- **Last Audited:** 2026-11-17 (Ref 2026-11)

## Factor Scores

| Factor | Score (/10) | Weight |
| --- | ---: | ---: |
| Privacy | 0 | 40% |
| Originality | 10 | 25% |
| Freshness | 3.5 | 20% |
| Grading | 7.3 | 15% |

## Primary-Source Evidence

### Privacy
> "We use the HealthBench and HealthBench Professional datasets that OpenAI publishes with simple-evals (commit 652c89d), pinned by checksum: 5,000 and 525 conversations."
- Source: https://artificialanalysis.ai/methodology/intelligence-benchmarking

### Originality
> "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."
- Source: https://arxiv.org/abs/2604.27470

### Freshness
> "arXiv:2604.27470 (cs) [Submitted on 30 Apr 2026]"
- Source: https://arxiv.org/abs/2604.27470

### Grading
> "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"
- Source: https://artificialanalysis.ai/methodology/intelligence-benchmarking

## Top Frontier Models on HealthBench

| Rank | Model | Creator | Percentile |
| ---: | --- | --- | ---: |
| #1 | [Claude Sonnet 5.5](https://benchaudit.org/models/claude-sonnet-5-5__max/) | Anthropic | 100.0p |
| #2 | [DeepSeek V4.1 Flash](https://benchaudit.org/models/deepseek-v4-1-flash__max/) | DeepSeek | 98.1p |
| #3 | [Claude Fable 5.2](https://benchaudit.org/models/claude-fable-5-2__max/) | Anthropic | 93.5p |
| #4 | [GPT-6.1 Sol](https://benchaudit.org/models/gpt-6-1-sol__max/) | OpenAI | 88.9p |
| #5 | [Claude Haiku 5.5](https://benchaudit.org/models/claude-haiku-5-5__max/) | Anthropic | 84.2p |
| #6 | [GPT-6 Astra](https://benchaudit.org/models/gpt-6-astra__max/) | OpenAI | 79.6p |
| #7 | [Claude Fable 5.1](https://benchaudit.org/models/claude-fable-5-1__max/) | Anthropic | 74.9p |
| #8 | [Claude Opus 5.5](https://benchaudit.org/models/claude-opus-5-5__max/) | Anthropic | 70.2p |
| #9 | [Grok 4.7](https://benchaudit.org/models/grok-4-7__xhigh/) | SpaceXAI | 65.6p |
| #10 | [GPT-6 Luna](https://benchaudit.org/models/gpt-6-luna__max/) | OpenAI | 61.0p |
