Health AI Benchmarks

Health Optimization Bench

Can a model keep up with fast-moving evidence in preventive medicine?

More about this benchmark

257 released tasks across 8 health-optimization subject suites, graded by a blind three-family panel.

Tasks
257
Snapshot
10 Sep 2026
Average panel agreement
83%

Scores come from runs by the benchmark's maintainers or independent researchers.

Results
16 from 15 models
Latest
Sep 2026
Published by
Health Optimization Bench
Released
Sep 2026
Sources
1

Where every published Health Optimization Bench score sits

16 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

16 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 5Anthropic70.9Benchmark teamSep 2026[1]
mean 0.709, ci 0.680-0.738, n 257
In the source
70.9
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Interval
68 to 73.8
Document
[1] Health Optimization Bench official results
2Claude Opus 5Anthropic69.3Benchmark teamSep 2026[1]
mean 0.693, ci 0.662-0.724, n 257
In the source
69.3
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Interval
66.2 to 72.4
Document
[1] Health Optimization Bench official results
3Grok 4.6SpaceX AI66.8Benchmark teamSep 2026[1]
mean 0.668, ci 0.638-0.698, n 257
In the source
66.8
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Interval
63.8 to 69.8
Document
[1] Health Optimization Bench official results
4GPT-5.6 Sol (max)OpenAI · max reasoning effort66.6Benchmark teamSep 2026[1]
mean 0.666, ci 0.634-0.696, n 257
In the source
66.6
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
How it was run
max reasoning effort
Reported by
official leaderboard
Interval
63.4 to 69.6
Document
[1] Health Optimization Bench official results
5GPT-5.6 Sol (high)OpenAI · high reasoning effort64.6Benchmark teamSep 2026[1]
mean 0.646, ci 0.615-0.676, n 257
In the source
64.6
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
How it was run
high reasoning effort
Reported by
official leaderboard
Interval
61.5 to 67.6
Document
[1] Health Optimization Bench official results
6Kimi K3Moonshot AI59.9Benchmark teamSep 2026[1]
mean 0.599, ci 0.565-0.633, n 257
In the source
59.9
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Interval
56.5 to 63.3
Document
[1] Health Optimization Bench official results
7Muse SparkMeta57.2Benchmark teamSep 2026[1]
mean 0.572, ci 0.539-0.605, n 257
In the source
57.2
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Interval
53.9 to 60.5
Document
[1] Health Optimization Bench official results
8Claude Fable 5.1Anthropic · declined 97 of 257 tasks47.3Benchmark teamSep 2026[1]
mean 0.473, ci 0.422-0.522, n 257
In the source
47.3
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Declined
declined 97 of 257 tasks
Interval
42.2 to 52.2
Document
[1] Health Optimization Bench official results
9Gemini 3.6Google39.7Benchmark teamSep 2026[1]
mean 0.397, ci 0.362-0.431, n 257
In the source
39.7
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Interval
36.2 to 43.1
Document
[1] Health Optimization Bench official results
10InklingThinking Machines35.6Benchmark teamSep 2026[1]
mean 0.356, ci 0.323-0.390, n 257
In the source
35.6
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Interval
32.3 to 39
Document
[1] Health Optimization Bench official results
11Claude Sonnet 5Anthropic34.6Benchmark teamSep 2026[1]
mean 0.346, ci 0.316-0.378, n 257
In the source
34.6
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Interval
31.6 to 37.8
Document
[1] Health Optimization Bench official results
12GLM 5.2Zhipu20.7Benchmark teamSep 2026[1]
mean 0.207, ci 0.181-0.234, n 257
In the source
20.7
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Interval
18.1 to 23.4
Document
[1] Health Optimization Bench official results
13MiniMax M3MiniMax18.3Benchmark teamSep 2026[1]
mean 0.183, ci 0.157-0.209, n 257
In the source
18.3
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Interval
15.7 to 20.9
Document
[1] Health Optimization Bench official results
14MAI ThinkingMicrosoft AI17.5Benchmark teamSep 2026[1]
mean 0.175, ci 0.151-0.200, n 257
In the source
17.5
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Interval
15.1 to 20
Document
[1] Health Optimization Bench official results
15Mistral Medium 3.5Mistral9.2Benchmark teamSep 2026[1]
mean 0.092, ci 0.074-0.111, n 257
In the source
9.2
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Interval
7.4 to 11.1
Document
[1] Health Optimization Bench official results
16Nemotron 3.5 LightningNVIDIA4.9Benchmark teamSep 2026[1]
mean 0.049, ci 0.036-0.062, n 257
In the source
4.9
Where to find it
tasksets/mb2ev/analysis.json, snapshot 2026-09-10
Reported by
official leaderboard
Interval
3.6 to 6.2
Document
[1] Health Optimization Bench official results
Sources

The 1 document behind these scores

  1. 1.Health Optimization Bench official resultsofficial leaderboard, Health Optimization Bench. Quotes from this document