Anthropic’s Claude Opus 5.5 Reduces Dashes, Extends Responses, Study Finds

Anthropic’s latest language model, Claude Opus 5.5, is showing a noticeable shift in its writing style, according to a new analysis that compared it with the prior version, Opus 5. The study observed a dramatic cut—about 95 percent—in the use of em dashes, a typographic feature often flagged as a hallmark of AI‑generated text. At the same time, the model’s answers have become lengthier, with sentences that are shorter but more numerous, and a tendency toward simpler vocabulary.

The reduction in em dashes appears intentional, as the punctuation can make AI output stand out to human reviewers and automated detectors alike. By trimming this stylistic quirk, Claude Opus 5.5 may aim to blend more seamlessly with human‑written content, a goal that aligns with Anthropic’s broader emphasis on “steerability” and safety. The analysis also noted that the newer model favors a more straightforward phrasing, avoiding the complex clauses that sometimes characterize earlier AI drafts.

While the model’s output has grown in overall length, the individual sentences are, on average, shorter than those produced by Opus 5. This pattern suggests a design choice to improve readability and reduce the cognitive load on users. Simpler wording and briefer sentences can make the information easier to digest, especially in contexts where quick comprehension is essential, such as technical support or news summarization.

Industry observers see these adjustments as part of a larger trend among AI developers to fine‑tune the “voice” of their systems. As large language models become more embedded in everyday applications—from customer service chatbots to content creation tools—reducing overtly artificial markers can help avoid suspicion and improve user trust. However, the increase in answer length raises questions about efficiency, particularly for users who prefer concise responses.

Anthropic has not released detailed commentary on the specific changes, but the findings underscore the company’s ongoing effort to balance naturalness with utility. Future updates may continue to refine the trade‑off between brevity and completeness, potentially offering configurable options that let users prioritize one over the other. For now, the data suggests Claude Opus 5.5 is moving toward a writing style that feels more human‑like while delivering more expansive answers, a development that could influence how AI assistants are evaluated and deployed across the tech landscape.