Skip to content

Benchmark

PhysicianBench

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.

Models measured
21
Scores recorded
42
Sources
3
IQ weight
1.69

Score distribution

36.3%
Mean
33.6%
Median
18.8%
Std dev
13.7%
10th percentile
63.2%
90th percentile
1.4%
Min
1.4%68.4%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.69 for chat models, measured over 21 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Opus 5.5adaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · 100 EHR tasks · max · five attempts/task· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.68.4%
2Claude Opus 5.5clinical-result-1830· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim68.4%
3Claude Fable 5.2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim64.4%
4Claude Sonnet 5.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim63.2%
5Claude Sonnet 5.5adaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · 100 EHR tasks · max · five attempts/task· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.63.2%
6Claude Sonnet 5.5clinical-result-1831· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim63.2%
7Claude Fable 5.1adaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · 100 EHR tasks · max · five attempts/task· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.61%
8Claude Fable 5.1clinical-result-1832· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim61%
9Claude Opus 5adaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · 100 EHR tasks · max · five attempts/task· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.57.6%
10Claude Opus 5clinical-result-1833· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim57.6%
11Claude Sonnet 5.5clinical-result-1834· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim56.4%
12Claude Sonnet 5.5clinical-result-1835· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim47.6%
13GPT 5.5clinical-result-167· Clinical Benchmarks data docs46.3%
14Claude Haiku 5.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim43%
15Claude Haiku 5.5adaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · 100 EHR tasks · max · five attempts/task· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.43%
16Claude Haiku 5.5clinical-result-2366· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim43%
17Claude Sonnet 5adaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · 100 EHR tasks · max · five attempts/task· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.37.4%
18Claude Sonnet 5clinical-result-1836· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim37.4%
19Claude Haiku 5.5adaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · 100 EHR tasks · xhigh · five attempts/task· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.35.8%
20Claude Haiku 5.5clinical-result-2367· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim35.8%
21GPT-6 Lunaadaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · 100 EHR tasks · max · five attempts/task· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.33.6%
22GPT-6 Lunaclinical-result-2373· Clinical Benchmarks data docs33.6%
23Claude Opus 4.6clinical-result-168· Clinical Benchmarks data docs31.7%
24Claude Haiku 5.5adaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · 100 EHR tasks · high · five attempts/task· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.31.6%
25Claude Haiku 5.5clinical-result-2368· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim31.6%
26Claude Sonnet 5.5clinical-result-1837· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim30%
27Claude Opus 4.7clinical-result-169· Clinical Benchmarks data docs29.3%
28GPT 5.4clinical-result-170· Clinical Benchmarks data docs27.7%
29Claude Sonnet 5.5clinical-result-1838· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim27.2%
30Claude Haiku 5.5adaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · 100 EHR tasks · medium · five attempts/task· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.25.2%
31Claude Haiku 5.5clinical-result-2369· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim25.2%
32Claude Sonnet 4.6clinical-result-171· Clinical Benchmarks data docs23%
33DeepSeek V4 Proclinical-result-1839· Clinical Benchmarks data docs18.7%
34Claude Haiku 5.5adaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · 100 EHR tasks · low · five attempts/task· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.17.8%
35Claude Haiku 5.5clinical-result-2370· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim17.8%
36Kimi K2.6clinical-result-172· Clinical Benchmarks data docs17%
37MiMo V2.5 Proclinical-result-1840· Clinical Benchmarks data docs16.7%
38Qwen3.6 Plusclinical-result-173· Clinical Benchmarks data docs13.7%
39MiniMax M2.7clinical-result-1841· Clinical Benchmarks data docs8.7%
40Gemini 3.1 Pro Previewclinical-result-174· Clinical Benchmarks data docs6%
41Grok 4.20clinical-result-295· Clinical Benchmarks data docs5.3%
42Claude Haiku 4.5clinical-result-2372· Clinical Benchmarks data docs· Reported by the model's developer. The source relaying it did not verify it. claim1.4%

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Clinical Benchmarks (clinicalbenchmarks.ai), CC BY 4.0; fractions converted to percentages; source documents retain their own terms (licence) · Data by LLM Stats