We saw this Opus 5.5 xHigh performing worse problem with Opus5 too where higher reasoning led to worse performance on FrontierCode because of scope creep.
Original titleWe saw this Opus 5.5 xHigh performing worse problem with Opus5 too where higher reasoning led to worse performance on FrontierCode becaus...
AISummary
Frontiercode penalizes unnecessary changes (second image) and models consistently perform worse at higher reasoning efforts
Source: wh · x.com