Users Report Abrupt Decline in Anthropic Opus Prose Quality

alvelda · x · 2026-08-12

Users on X have reported a noticeable decline in the prose generation quality of Anthropic's Opus models. Tasks that previously took one turn to achieve refined grammar and style now require up to eight iterations, with the model struggling to write a decent introductory paragraph for a paper. This sparks discussions on potential capability regressions following silent model updates.

Original post →

More from Models

Models channel →