A significant update has been made to Anthropic’s Claude Opus, a popular AI model used for writing and coding tasks. The new version, Opus 5.5, has undergone changes that make it less identifiable as AI-generated content, according to an analysis by Arena, an AI benchmarking tool. This shift is notable, as previous versions of Opus were often criticized for their reliance on obvious AI writing patterns.
One of the most striking changes in Opus 5.5 is its reduced use of em dashes. In fact, the new model uses a staggering 95% fewer em dashes than its predecessor, Opus 5. For those unfamiliar with linguistic nuances, an em dash (—) is a punctuation mark often used to set off parenthetical phrases or indicate a pause in thought. AI-generated content frequently employs em dashes as a way to create a sense of fluidity and complexity, making it easier to detect. By reducing their use, Anthropic has made it more difficult for humans to identify Opus 5.5’s writing as artificial.
The analysis also reveals that Opus 5.5 produces shorter sentences on average, averaging 10.03 words compared to the 12.14 words in Opus 5. This simplification of language makes it less likely for readers to suspect AI-generated content. However, this shift comes with a trade-off: Opus 5.5’s answers are becoming longer overall, with an increase from 453 to 481 words on average. While this may not be a concern for some users, it could potentially impact those who rely on Claude for concise and efficient responses.
The changes in Opus 5.5 suggest that Anthropic is making a concerted effort to refine its AI model’s writing capabilities. By eliminating obvious AI patterns and simplifying language, the company aims to create a more natural and engaging experience for users. As AI models continue to evolve, it’s essential for developers to prioritize transparency and subtlety in their creations.
For those concerned about identifying AI-generated content, Opus 5.5’s reduced use of em dashes is an encouraging sign. Nevertheless, readers should remain vigilant and familiarize themselves with the characteristics of AI writing patterns to avoid being misled by overly verbose or simplistic language. By staying informed and adaptable, individuals can better navigate the rapidly evolving landscape of artificial intelligence and its applications.
Source: Bleeping Computer — 2026-09-26