Anthropic's Claude 5.5 Writing Quality Declines Despite Model Improvements

Jackson Kernion, an engineer at Anthropic, reported that the writing quality of the Claude AI model has stalled or worsened with the release of Opus 5.5, despite improvements in math, code, and reasoning capabilities. Kernion expressed dissatisfaction with the writing quality, stating he hasn't been 'as happy about a model's writing since Opus 4.6', which he described as the last good writing model. He noted that Claude's writing now features 'odd phrasing' and 'overly-dense info dumps', which he attributes to the model's adaptation to 'LLM psychology'. Kernion acknowledged the challenges in solving these writing issues, emphasizing that they will continue to make improvements. This follows previous discussions on the performance of Claude models, particularly the contrast between Opus 4.6 and 5.5.