A 14B model beats every frontier giant at ideological evenhandedness

Claim

A new political-consistency benchmark found a specially trained 14-billion-parameter model beating every frontier giant at ideological evenhandedness, while a benchmark treating an obscure board game as a proxy for learning on the job saw a new 39.7% state of the art still trailing expert humans.

Reported metric: 39.7% vs expert humans

Reporting date: 2026-07-11 (not necessarily the event date)

Sources

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

• Welcome to July 11, 2026 (theinnermostloop.substack.com)