Ethan Mollick: model personality matters beyond benchmarks — a bad release doesn't spell decline

eldonredwards · x · 2026-11-18

Ethan Mollick argues that model "personality" matters in ways benchmarks can't capture. He recalls Anthropic's rough patch: from around Opus 4.7 to 5, models "didn't feel Claude-y anymore," more like a watered-down Fable (he suspects teacher-model distillation); Opus 5.5 finally feels like working with ol' Claude again.

On the recent panning of OpenAI's 6.2, he cautions that people very close to AI news on X tend to extrapolate too much from each release: Anthropic stumbled before Opus 5.5, Google had a long dry stretch — "6.2 is not terrific but that doesn't tell us much."

Original post →

More from AGI Musings

AGI Musings channel →