HealthBench Professional
We show this table for reference; we do not rank on it.
An open benchmark for clinician-facing model responses across care consult, writing and documentation, and medical research tasks.
Benchmark score on HealthBench Professional — November 5, 2026
We compile the HealthBench Professional rows from benchmark-owner or independent runs and provider self-reports. Claude Sonnet 5.5 leads the table at 69.2%, followed by Claude Opus 5.5 (65.6%) and Ling 3.1 Flash (65.3%). We do not use these results to rank models overall.
Claude Sonnet 5.5
Anthropic
Claude Opus 5.5
Anthropic
Ling 3.1 Flash
InclusionAI
18 modelsKnowledgeCurrentDisplay onlyUpdated November 5, 2026
Benchmark score table (18 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Sonnet 5.5Anthropic | 69.2% | Not reported | Closed |
| 2 | Claude Opus 5.5Anthropic | 65.6% | Not reported | Closed |
| 3 | Ling 3.1 FlashInclusionAI | 65.3% | Not reported | Open |
| 4 | GPT-6 AstraOpenAI | 64.7% | Not reported | Closed |
| 5 | GPT-6.1 SolOpenAI | 64.2% | Not reported | Closed |
| 6 | GPT-6.1 TerraOpenAI | 62.7% | Not reported | Closed |
| 7 | Muse Wonder FullMeta | 62.3% | Not reported | Closed |
| 8 | GPT-6 SolOpenAI | 60.8% | Not reported | Closed |
| 9 | GPT-6 LunaOpenAI | 60.8% | Not reported | Closed |
| 10 | GPT-5.6 SolOpenAI | 60.5% | Not reported | Closed |
| 11 | Muse WonderMeta | 60.5% | Not reported | Closed |
| 12 | Claude Opus 5Anthropic | 59.8% | Not reported | Closed |
| 13 | Muse Spark 1.1Meta | 59.3% | Not reported | Closed |
| 14 | GPT-5.6 TerraOpenAI | 57.7% | Not reported | Closed |
| 15 | Grok 4.7xAI | 56.7% | Not reported | Closed |
| 16 | GPT-5.6 LunaOpenAI | 55.7% | Not reported | Closed |
| 17 | Ling 3.0 Flash SanteInclusionAI | 54.3% | Not reported | Open |
| 18 | GPT-5.4OpenAI | 48.1% | Not reported | Closed |
HealthBench Professional (length-adjusted, Artificial Analysis)
Artificial Analysis evaluates 525 conversations with GPT-5.4 (low) as judge and 4 repeats per task. Length-adjusted rubric score is reported in percentage points. Length adjustment uses 1.47 points per 500 answer characters above or below 2,000; reasoning is excluded. Effort and Default Fallback labels identify the evaluated configuration. These rows remain separate from provider runs and outside overall and category rankings.
| Source rank | Source model or configuration | Length-adjusted rubric score | Source setup | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra (Max)OpenAI | 66.0% | MaxOriginal result | Not reported | Closed |
| 2 | GPT-6.1 Sol (Max)OpenAI | 65.1% | MaxOriginal result | Not reported | Closed |
| 3 | Claude Sonnet 5.5 (Max, Default Fallback)Anthropic | 63.0% | Max, Default FallbackOriginal result | Not reported | Closed |
| 4 | GPT-6 Sol (Max)OpenAI | 62.3% | MaxOriginal result | Not reported | Closed |
| 5 | GPT-6 Luna (Max)OpenAI | 61.9% | MaxOriginal result | Not reported | Closed |
| 6 | Claude Opus 5.5 (Max, Default Fallback)Anthropic | 59.9% | Max, Default FallbackOriginal result | Not reported | Closed |
| 7 | Claude Haiku 5.5 (Max)Anthropic | 58.4% | MaxOriginal result | Not reported | Closed |
| 8 | Claude Fable 5.1 (Max, Default Fallback)Anthropic | 57.8% | Max, Default FallbackOriginal result | Not reported | Closed |
| 9 | Grok 4.7 (Xhigh)xAI | 55.8% | XhighOriginal result | Not reported | Closed |
| 10 | Gemini 4 Argon (High)Google | 55.1% | HighOriginal result | Not reported | Closed |
| 11 | Kimi K3 (Max)Moonshot AI | 53.2% | MaxOriginal result | Not reported | Open |
| 12 | GLM-5.3 (Max)Z.AI | 52.8% | MaxOriginal result | Not reported | Open |
| 13 | Gemini 3.8 Flash (High)Google | 51.7% | HighOriginal result | Not reported | Closed |
| 14 | DeepSeek V4.1 Flash (Max)DeepSeek | 50.7% | MaxOriginal result | Not reported | Open |
Among the reported HealthBench Professional rows, Claude Sonnet 5.5 is first at 69.2%. The third row is 3.9 points behind. The broader top-10 range is 8.7 points, so many of the published results sit in a relatively narrow band.
18 models have been evaluated on HealthBench Professional. The benchmark falls in the Knowledge category. HealthBench Professional is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About HealthBench Professional
Year
2026
Tasks
Clinician chat tasks
Format
Rubric-graded open-ended responses
Difficulty
Professional clinical workflows
HealthBench Professional uses physician-authored clinician conversations and physician-written rubrics to evaluate model responses on difficult clinical tasks. BenchLM tracks it as a display-only health benchmark because some source rows include product harnesses rather than raw base-model capability.
Freshness and provenance
Version
HealthBench Professional 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does HealthBench Professional measure?
An open benchmark for clinician-facing model responses across care consult, writing and documentation, and medical research tasks.
Which model scores highest on HealthBench Professional?
Claude Sonnet 5.5 by Anthropic currently leads with a score of 69.2% on HealthBench Professional.
How many models are evaluated on HealthBench Professional?
18 AI models have published results on HealthBench Professional in the BenchLM catalog.
Compare top models on HealthBench Professional
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.