Development

A specialist model more than doubles the generalist on hard health questions

development2026-03-04

Claim

KOS-1 Lite scored 46.6% on HealthBench Hard against Claude Opus 4.6 at 20.4%, at a fraction of the serving cost.

Reported metric: 46.6% vs 20.4%

Reporting date: 2026-03-04 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Welcome to March 4, 2026 (theinnermostloop.substack.com)

A counterweight to the generalism thesis: on one domain, the specialist wins by more than double.

Knowledge graphopen in browser →

Reported in

Welcome to March 4, 2026

Evidence for

Supersedes

Law firms brand themselves Claude-native

Referenced by

Welcome to March 4, 2026

Cite or correct this record

Citation formatCitation text

A specialist model more than doubles the generalist on hard health questions
Innermost Loop KB — secondary record
Reporting date: 2026-03-04
KB permalink: https://nicholsn.github.io/innermost-loop-kb/developments/2026-03-04-specialist-doubles-the-generalist-on-health
Claim: KOS-1 Lite scored 46.6% on HealthBench Hard against Claude Opus 4.6 at 20.4%, at a fraction of the serving cost.
Attributed sources:
- Welcome to March 4, 2026 — https://theinnermostloop.substack.com/p/welcome-to-march-4-2026
This citation does not imply independent verification of linked-source claims.

Suggest a correction — opens a GitHub draft for you to review and submit.