Viral BridgeBench Post Claims Claude Opus 4.6 Was ‘Nerfed,’ Critics Call It Bad Science

BridgeMind AI claimed Anthropic's Claude Opus 4.6 was secretly downgraded after its hallucination benchmark score dropped from 83.3% to 68.3%. Critics, including computer scientist Paul Calcraft, called the comparison flawed because the original score was based on 6 tasks and the retest on 30 tasks. On the 6 overlapping tasks, performance was nearly identical (87.6% vs 85.4%), with the swing due to a single fabrication within normal variance. Broader complaints about Claude Opus 4.6's quality may stem from Anthropic's adaptive thinking controls, which prioritize efficiency. The benchmark does not prove a deliberate downgrade.