HealthBench ProfessionalBrowse 296

HealthBench Professional

Can the model produce useful work in a professional knowledge-work setting?

Version not specifiedExact reported variant
Knowledge Work8 ranked models10 reported values3 reportsHigher is better

Top rankings

One row per model, using its best reported score across effort settings.

8
RankModelBest scoreBest reported setting
1Claude Mythos PreviewAnthropic64.7%Reported configurationAnthropic report1 value · 1 reportJun 9, 2026 · source
2Claude Fable 5Anthropic60.9%Reported configurationOpenAI report1 value · 1 reportJul 9, 2026 · source
3GPT-5.6 SolOpenAI60.5%Reported configurationOpenAI report1 value · 1 reportJul 9, 2026 · source
4GPT-5.6 TerraOpenAI57.7%Reported configurationOpenAI report1 value · 1 reportJul 9, 2026 · source
5Claude Opus 4.8Anthropic56.9%Reported configurationAnthropic report2 values · 2 reportsJun 9, 2026 · source
6Grok 4.7SpaceXAI56.7%xhigh effortSpaceXAI report1 value · 1 reportSep 21, 2026 · source
7GPT-5.6 LunaOpenAI55.7%Reported configurationOpenAI report1 value · 1 reportJul 9, 2026 · source
8GPT-5.5OpenAI51.8%Reported configurationAnthropic report2 values · 2 reportsJun 9, 2026 · source

Effort curve

Every sourced cost-linked effort value for this exact version. Lines connect complete sweeps only.

0
No cost-linked effort sweep for this version.
Definition and comparison boundarypublic methodology

Can the model produce useful work in a professional knowledge-work setting?

Professional analysis, document, finance, office, or domain-specific tasks. Higher is better. The value is the percentage reported in this lab's table.

Harness, tools, prompts, attempt count, and benchmark version can materially change the result. Compare within one reporting context.

SpaceXAI reports 56.7% for Grok 4.7 at xhigh; the table does not print the benchmark harness or task revision.

Method / source