Health Optimization Bench: published results

Arcophos · 257 tasks across eight subject suites · index updated September 28, 2026

Claude Fable 5 has the highest indexed numerical score on Health Optimization Bench, 70.9 as of 2026-09, per Health Optimization Bench — subject suites results. Questions in eight areas of preventive and optimization medicine, grounded in primary evidence and scored against task-specific rubrics. The current main ranking covers 257 released tasks, separate from the 89-task incretin therapeutics evidence suite.

This benchmark has a dedicated full leaderboard, with methodology and per-model pages, at healthoptimizationbench.com. Every published row in the index appears below.

Published results

#	model		score	as of
1		Claude Fable 5 Anthropic Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 68.0–73.8. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. via Health Optimization Bench — subject suites results	70.9	2026-09
2		Claude Opus 5 Anthropic Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 66.2–72.4. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. via Health Optimization Bench — subject suites results	69.3	2026-09
3		Grok 4.6 xAI Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 63.8–69.8. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. via Health Optimization Bench — subject suites results	66.8	2026-09
4		GPT-5.6 Sol (max) OpenAI Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 63.4–69.6. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. Maximum reasoning effort. via Health Optimization Bench — subject suites results	66.6	2026-09
5		GPT-5.6 Sol (high) OpenAI Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 61.5–67.6. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. High reasoning effort. via Health Optimization Bench — subject suites results	64.6	2026-09
6		Kimi K3 Moonshot AI Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 56.5–63.3. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. via Health Optimization Bench — subject suites results	59.9	2026-09
7		Muse Spark Meta Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 53.9–60.5. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. via Health Optimization Bench — subject suites results	57.2	2026-09
8		Claude Fable 5.1 Anthropic Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 42.2–52.2. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. Vendor safeguards declined 97/257 tasks, scored with no credit; mean over answered tasks is 75.9. via Health Optimization Bench — subject suites results	47.3	2026-09
9		Gemini 3.6 Google Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 36.2–43.1. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. via Health Optimization Bench — subject suites results	39.7	2026-09
10	TM	Inkling Thinking Machines Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 32.3–39.0. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. via Health Optimization Bench — subject suites results	35.6	2026-09
11		Claude Sonnet 5 Anthropic Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 31.6–37.8. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. via Health Optimization Bench — subject suites results	34.6	2026-09
12		GLM 5.2 Zhipu Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 18.1–23.4. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. via Health Optimization Bench — subject suites results	20.7	2026-09
13		MiniMax M3 MiniMax Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 15.7–20.9. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. via Health Optimization Bench — subject suites results	18.3	2026-09
14		MAI Thinking Microsoft AI Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 15.1–20.0. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. via Health Optimization Bench — subject suites results	17.5	2026-09
15		Mistral Medium 3.5 Mistral Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 7.4–11.1. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. via Health Optimization Bench — subject suites results	9.2	2026-09
16		Nemotron 3.5 Lightning NVIDIA Subject suites release set: 257 tasks across eight subjects; one answer per task, no tools; blind cross-family grading. 95% bootstrap CI 3.6–6.2. Harness snapshot September 10, 2026; not the separate 89-task incretin ranking. via Health Optimization Bench — subject suites results	4.9	2026-09

Scores preserve their source precision, with any scale conversion documented (independently run). The main ranking moved from the 89-task incretin suite to 257 subject-suite tasks. This page shows only the 257-task September 10 snapshot; the two task sets must not be pooled. Three grader families vote, with split decisions escalated to a fourth, and the authoring family excluded. Fable 5.1 safeguards declined 97 tasks; those receive no credit. The line under each model names the document its score was read from; rows marked source pending are still awaiting a documented first-party source. Full citations are on the sources page.

About the benchmark

The rest of the field is on the index, and how sources qualify is on the methodology page. Model names in the table link to cross-benchmark pages.