Health Optimization Bench: incretin evidence synthesis
Can a model keep up with the evidence on GLP-1 and other incretin drugs?
More about this benchmark
Hard, freshness-dependent questions in preventive and optimization medicine, written against primary sources and graded blind by a panel.
Frontier models on hard, freshness-dependent questions in preventive and optimization medicine. Every task is written against a primary source, audited by model families that did not author it, and scored blind by a panel of independent families. The v1 release set covers incretin therapeutics evidence.
- Released
- Aug 2026
- Snapshot
- 10 Sep 2026
- Average panel agreement
- 87%
Reading these numbers
Cross-family authoring with blind three-family panel grading; the authoring family never grades its own task, and 95 percent bootstrap confidence intervals accompany every score on the site.
Scores come from runs by the benchmark's maintainers or independent researchers.
- Results
- 18 from 17 models
- Latest
- Sep 2026
- Published by
- Health Optimization Bench
- Index weight
- 0.226, supporting, near its ceiling
- Sources
- 1
Counts toward the Health Evals Index as a supporting benchmark, at weight 0.226: prevention and optimization questions, one area of personal health. It is near its ceiling, with a leader at 84.9 out of 100, so it counts for less.
Where every published Health Optimization Bench: incretin evidence synthesis score sits
18 scores, one column each, highest first, on the benchmark's own scale. Higher is better.
Each column is the model's lab colour. The table below says who ran each score.
Every published score, ranked within this benchmark only
18 scores. Higher is better. Where the source gave a range, the whisker shows it. Open any row to read the sentence the score came from.
1Claude Fable 5.1Anthropic · declined 1 of 89 tasks84.9Benchmark teamSep 2026[1]
mean 0.849, ci 0.801-0.893, n 89
2Claude Fable 5Anthropic83.8Benchmark teamSep 2026[1]
mean 0.838, ci 0.791-0.881, n 89
3GPT-6 Astra (max)OpenAI · max reasoning effort82.2Benchmark teamSep 2026[1]
mean 0.822, ci 0.777-0.864, n 89
4Grok 4.6SpaceX AI81.3Benchmark teamSep 2026[1]
mean 0.813, ci 0.770-0.853, n 89
5Claude Opus 5Anthropic78.3Benchmark teamSep 2026[1]
mean 0.783, ci 0.729-0.834, n 89
6GPT-5.6 Sol (max)OpenAI · max reasoning effort77.9Benchmark teamSep 2026[1]
mean 0.779, ci 0.733-0.822, n 89
7Kimi K3Moonshot AI77.8Benchmark teamSep 2026[1]
mean 0.778, ci 0.730-0.824, n 89
8GPT-5.6 Sol (high)OpenAI · high reasoning effort77.7Benchmark teamSep 2026[1]
mean 0.777, ci 0.731-0.820, n 89
9Muse SparkMeta68.4Benchmark teamSep 2026[1]
mean 0.684, ci 0.628-0.739, n 89
10Gemini 3.8 FlashGoogle67.8Benchmark teamSep 2026[1]
mean 0.678, ci 0.617-0.736, n 89
11Gemini 3.6Google57.9Benchmark teamSep 2026[1]
mean 0.579, ci 0.512-0.643, n 89
12InklingThinking Machines53.9Benchmark teamSep 2026[1]
mean 0.539, ci 0.477-0.601, n 89
13Claude Sonnet 5Anthropic50.1Benchmark teamSep 2026[1]
mean 0.501, ci 0.443-0.561, n 89
14MiniMax M3MiniMax35.5Benchmark teamSep 2026[1]
mean 0.355, ci 0.296-0.416, n 89
15MAI ThinkingMicrosoft AI33.0Benchmark teamSep 2026[1]
mean 0.330, ci 0.268-0.393, n 89
16GLM 5.2Zhipu29.2Benchmark teamSep 2026[1]
mean 0.292, ci 0.239-0.349, n 89
17Mistral Medium 3.5Mistral14.4Benchmark teamSep 2026[1]
mean 0.144, ci 0.106-0.185, n 89
18Nemotron 3.5 LightningNVIDIA7.8Benchmark teamSep 2026[1]
mean 0.078, ci 0.047-0.113, n 89
The 1 document behind these scores
- 1.Health Optimization Bench official resultsofficial leaderboard, Health Optimization Bench. Quotes from this document