HealthBench length-adjusted score (HealthBench (length-adjusted))
We show this table for reference; we do not rank on it.
HealthBench score after applying a verbosity penalty to model responses.
Benchmark score on HealthBench (length-adjusted) — November 13, 2026
We compile the HealthBench (length-adjusted) rows from provider self-reports. Claude Sonnet 5.5 leads the table at 65.4%, followed by Claude Opus 5.5 (60.6%) and GPT-6.1 Sol (58.5%). We do not use these results to rank models overall.
Claude Sonnet 5.5
Anthropic
Reported by AnthropicClaude Opus 5.5
Anthropic
Reported by AnthropicGPT-6.1 Sol
OpenAI
Reported by OpenAI8 modelsKnowledgeCurrentDisplay onlyUpdated November 13, 2026
Benchmark score table (8 models)
All 8 results on this page were reported by the labs that built the models.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | Claude Sonnet 5.5AnthropicReported by Anthropic | 65.4% | Reported by Anthropic | Not reported | Closed |
| 2 | Claude Opus 5.5AnthropicReported by Anthropic | 60.6% | Reported by Anthropic | Not reported | Closed |
| 3 | GPT-6.1 SolOpenAIReported by OpenAI | 58.5% | Reported by OpenAI | Not reported | Closed |
| 4 | GPT-6 AstraOpenAIReported by OpenAI | 58.3% | Reported by OpenAI | Not reported | Closed |
| 5 | Claude Opus 5AnthropicReported by Anthropic | 57.8% | Reported by Anthropic | Not reported | Closed |
| 6 | GPT-6.1 TerraOpenAIReported by OpenAI | 55.1% | Reported by OpenAI | Not reported | Closed |
| 7 | GPT-6 LunaOpenAIReported by OpenAI | 54.5% | Reported by OpenAI | Not reported | Closed |
| 8 | GPT-6 SolOpenAIReported by OpenAI | 53.2% | Reported by OpenAI | Not reported | Closed |
- Lab-reported
- Benchmark-run
- Independent
HealthBench (length-adjusted, Artificial Analysis)
Artificial Analysis evaluates 5,000 conversations with GPT-4.1 as judge and 1 run per task. Length-adjusted rubric score is reported in percentage points. Length adjustment uses 2.99 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 | Claude Sonnet 5.5 (Max, Default Fallback)Anthropic | 62.1% | Max, Default FallbackOriginal result | Not reported | Closed |
| 2 | DeepSeek V4.1 Flash (Max)DeepSeek | 60.9% | MaxOriginal result | Not reported | Open |
| 3 | Claude Fable 5.2 (Max, Default Fallback)Anthropic | 59.6% | Max, Default FallbackOriginal result | Not reported | Closed |
| 4 | GPT-6.1 Sol (Max)OpenAI | 58.5% | MaxOriginal result | Not reported | Closed |
| 5 | Claude Haiku 5.5 (Max)Anthropic | 58.5% | MaxOriginal result | Not reported | Closed |
| 6 | GPT-6 Astra (Max)OpenAI | 58.2% | MaxOriginal result | Not reported | Closed |
| 7 | Claude Fable 5.1 (Max, Default Fallback)Anthropic | 57.2% | Max, Default FallbackOriginal result | Not reported | Closed |
| 8 | Claude Opus 5.5 (Max, Default Fallback)Anthropic | 57.0% | Max, Default FallbackOriginal result | Not reported | Closed |
| 9 | Grok 4.7 (Xhigh)xAI | 56.0% | XhighOriginal result | Not reported | Closed |
| 10 | Kimi K3 (Max)Moonshot AI | 54.7% | MaxOriginal result | Not reported | Open |
| 11 | GPT-6 Luna (Max)OpenAI | 54.6% | MaxOriginal result | Not reported | Closed |
| 12 | GLM-5.3 (Max)Z.AI | 54.5% | MaxOriginal result | Not reported | Open |
| 13 | GPT-6 Sol (Max)OpenAI | 54.2% | MaxOriginal result | Not reported | Closed |
| 14 | Gemini 4 Argon (High)Google | 49.1% | HighOriginal result | Not reported | Closed |
| 15 | Gemini 3.8 Flash (High)Google | 48.4% | HighOriginal result | Not reported | Closed |
Among the reported HealthBench (length-adjusted) rows, Claude Sonnet 5.5 is first at 65.4%. The third row is 6.9 points behind. The broader top-10 range is 12.2 points, so the table still separates the published systems.
8 models have been evaluated on HealthBench (length-adjusted). The benchmark falls in the Knowledge category. HealthBench (length-adjusted) is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About HealthBench (length-adjusted)
Year
2025
Tasks
5,000 multi-turn patient conversations
Format
Length-adjusted rubric score
Difficulty
Realistic healthcare conversations
Length-adjusted scores remain separate from raw rubric scores. Provider reports retain their own trial settings and source citations. Artificial Analysis runs HealthBench once per task with GPT-4.1 as judge; those independent results appear in a separate table.
Freshness and provenance
Version
HealthBench (length-adjusted) 2025
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 (length-adjusted) measure?
HealthBench score after applying a verbosity penalty to model responses.
Which model scores highest on HealthBench (length-adjusted)?
Claude Sonnet 5.5 by Anthropic currently leads with a score of 65.4% on HealthBench (length-adjusted).
How many models are evaluated on HealthBench (length-adjusted)?
8 AI models have published results on HealthBench (length-adjusted) in the BenchLM catalog.
Compare top models on HealthBench (length-adjusted)
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 7,000+ subscribers.
One email each week. Unsubscribe anytime.